petersson inner product of theta series...cette thèse se penche sur plusieurs aspects du produit...
TRANSCRIPT
Petersson Inner Product of Theta Series
Author: Nicolas Simard
A thesis submitted to McGill Universityin partial fullment of the requirements of the
degree of Doctor of Philosophy
Department of Mathematics and Statistics
McGill University
Montreal
December 2017
Supervisor: Henri Darmon
©Nicolas Simard, 2017
Acknowledgements
Of course, this project wouldn't have been possible without the help and patience of
Prof. Darmon. I would like to thank him for his numerous advice, his thoughts on the
project and for the time he took to discuss with me over the years. His mathematical
insight is remarkable and I sometimes feel that I spent the past few years understanding
and proving that his intuition was right! I would also like to thank the National Science
and Engineering Research Council (NSERC) and the McGill Mathematics and Statistics
Department for their nancial support. Enn, j'aimerais remercier ma femme Andréanne
et ma lle Éliane d'apporter autant de bonheur dans ma vie. Être en leur présence me
remplit d'amour et me motive à mener à terme ce projet qu'est le Doctorat.
ii
Abstract
In this thesis, we are interested in many aspects of the Petersson inner product of theta
series attached to imaginary quadratic elds. In the rst part, we nd closed formulas for
the Petersson norm of theta series attached to unramied Hecke characters of varying
innity type. Using these, we nd formulas for the Petersson inner product of theta series
attached to ideals. We then see how this generalizes a remark made by Stark in [Sta75]
about the connection between the Petersson norm of the theta series attached to certain
Artin representations and the value at s = 1 of the corresponding Artin L-functions. In
the second part of the thesis, we show that the Petersson inner product of theta series can
be p-adically interpolated when p splits in the imaginary quadratic eld. We then evaluate
the constructed p-adic analytic function outside the range of interpolation and show how
the expression obtained can be seen as a p-adic analogue of the corresponding classical
expression. In the third and last part of the thesis, we present some computations to give
examples, illustrate the theory and support conjectures made in the text.
iii
Abrégé
Cette thèse se penche sur plusieurs aspects du produit scalaire de Petersson des foncti-
ons thêta associées aux corps quadratiques imaginaires. Dans la première partie, on donnera
des formules explicites pour le produit scalaire de Petersson des fonctions thêta associées à
une famille de charactères de Hecke. Ces formules serviront ensuite à trouver des formules
pour le produit scalaire des fonctions thêta associées à des idéaux fractionaires. On verra
ensuite comment ces formules permettent de généraliser une remarque faite par Stark dans
[Sta75] à propos d'une relation entre le produit scalaire de Petersson des fonctions thêta et
la valeur en s = 1 de certaines fonctions L de Artin. Dans la seconde partie de la thèse, on
montrera que le produit scalaire de Petersson d'une certaine collection de fonctions thêta
s'interpole p-adiquement pour certain nombres premiers p. On verra ensuite qu'en évaluant
la fonction p-adique obtenue à un entier non interpolé, on obtient une expression analogue
à l'expression classique correspondante. Enn, la dernière partie de la thèse contient plu-
sieurs exemples de calculs qui illustrent la théorie et supportent les conjectures faites dans
le texte.
iv
TABLE OF CONTENTS
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Abrégé . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
I Complex formulas 7
1 Modular forms over C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1 The spaces GL+, H, LΓ and Y (Γ) . . . . . . . . . . . . . . . . . . . . . 81.2 Weight and q-expansions . . . . . . . . . . . . . . . . . . . . . . . . . . . 111.3 Modular forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131.4 Holomorphic and C∞ Eisenstein series . . . . . . . . . . . . . . . . . . . 151.5 Operators on Mk(Γ0(N), χ) . . . . . . . . . . . . . . . . . . . . . . . . . 211.6 Petersson inner product and Atkin-Lehner Theory . . . . . . . . . . . . 271.7 Dierential operators on modular forms . . . . . . . . . . . . . . . . . . 291.8 The L-function of modular forms . . . . . . . . . . . . . . . . . . . . . . 31
2 Computing the Petersson inner product of cusp forms . . . . . . . . . . . . . . 33
2.1 The Rankin-Selberg method . . . . . . . . . . . . . . . . . . . . . . . . . 332.2 Rankin- Selberg convolution and symmetric square L-functions . . . . . 352.3 F Petersson norm of newforms and special values of their symmetric
square L-functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382.4 F Petersson norm of newforms with real Fourier coecients . . . . . . . 41
3 Complex multiplication and modular forms over C . . . . . . . . . . . . . . . . 43
3.1 CM points in H, L and Y (SL2(Z)) . . . . . . . . . . . . . . . . . . . . . 433.2 F CM values of modular functions . . . . . . . . . . . . . . . . . . . . . 443.3 F Siegel units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453.4 F CM values of modular forms . . . . . . . . . . . . . . . . . . . . . . . 463.5 F CM values of nearly holomorphic modular forms . . . . . . . . . . . . 47
v
4 Petersson inner product of theta functions . . . . . . . . . . . . . . . . . . . . . 48
4.1 F Hecke characters of type A0 with trivial conductor . . . . . . . . . . . 484.2 F Theta functions attached to imaginary quadratic elds . . . . . . . . 504.3 F Petersson inner product of theta functions attached to imaginary
quadratic elds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5 Petersson norm of weight one theta functions . . . . . . . . . . . . . . . . . . . 62
5.1 F Petersson norm of theta functions and Siegel units . . . . . . . . . . . 625.2 F On Stark's observation . . . . . . . . . . . . . . . . . . . . . . . . . . 625.3 F On generalisations of Stark's observation . . . . . . . . . . . . . . . . 64
II p-adic Interpolation 68
6 p-adic modular forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6.1 Towards an algebraic theory of modular forms . . . . . . . . . . . . . . . 696.2 Test objects and trivialized elliptic curves . . . . . . . . . . . . . . . . . 746.3 Weight and Tate curves . . . . . . . . . . . . . . . . . . . . . . . . . . . 766.4 Algebraic and generalized p-adic modular forms . . . . . . . . . . . . . . 796.5 Comparison between modular forms . . . . . . . . . . . . . . . . . . . . 826.6 p-adic Eisenstein series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 846.7 Changing the level at p . . . . . . . . . . . . . . . . . . . . . . . . . . . . 856.8 Operators on generalized p-adic modular forms . . . . . . . . . . . . . . 876.9 Theta operator on generalized p-adic modular forms . . . . . . . . . . . 88
7 Complex multiplication from an algebraic point of view . . . . . . . . . . . . . 90
7.1 F CM values of algebraic modular forms . . . . . . . . . . . . . . . . . . 917.2 CM values of p-adic modular forms . . . . . . . . . . . . . . . . . . . . . 93
8 p-adic interpolation of Petersson inner product of theta series . . . . . . . . . . 95
8.1 Review of p-adic integration theory . . . . . . . . . . . . . . . . . . . . . 958.2 Construction of measures with values in the ring of generalized p-adic
modular forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1028.3 Construction of a measure with values in Cp . . . . . . . . . . . . . . . . 1068.4 p-adic interpolation of Petersson inner product of theta series . . . . . . 1108.5 p-adic analogue of Kronecker's First Limit Formula . . . . . . . . . . . . 112
vi
III Computations 118
9 Petersson inner product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
9.1 Petersson norm of ∆ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1209.2 Petersson norm of ∆5(τ) = (η(τ)η(5τ))4 . . . . . . . . . . . . . . . . . . 1219.3 Petersson norm of ∆7(τ) = (η(τ)η(7τ))3 . . . . . . . . . . . . . . . . . . 122
10 Complex multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
10.1 Singular moduli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12410.2 Siegel units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12510.3 CM values of modular forms: classically and algebraically . . . . . . . . 126
11 Petersson inner product of theta series . . . . . . . . . . . . . . . . . . . . . . . 129
11.1 Hecke characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12911.2 Theta functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13011.3 Petersson inner product of theta functions . . . . . . . . . . . . . . . . . 131
12 Stark's observation on weight one theta functions . . . . . . . . . . . . . . . . . 136
12.1 Stark's original observation in the eld Q(√−23) . . . . . . . . . . . . . 136
12.2 Generalizing Stark's observation to class number 3 quadratic elds . . . 13612.3 Generalizing Stark's observation to other quadratic elds . . . . . . . . . 138
13 Computational experiments and conjectures . . . . . . . . . . . . . . . . . . . . 140
13.1 Gram matrix of the Petersson inner product on the space of theta series 14013.2 Invariant theta series attached to imaginary quadratic elds . . . . . . . 143
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
vii
Introduction
L-functions are central objects in number theory. They can be attached to a variety of
mathematical objects, like Dirichlet characters, modular forms, algebraic varieties, Galois
representations, etc. and contain a surprisingly large amount of information about the
object to which they are attached.
For example, take the simplest possible L-function, the Riemann zeta function, which
is dened for <(s) > 1 by the Dirichlet series
ζ(s) =
∞∑n=1
1
ns.
The simple fact that it has a pole at s = 1 implies that there are innitely many primes.
With more work, one can also obtain the prime number theorem about the distribution of
primes using the fact that ζ(s) does not vanish close to the line <(s) = 1. It is also known
that one could obtain the best possible error bound on the Prime Number Theorem by
knowing the Riemann Hypothesis, which conjectures that the non-trivial zeroes of ζ(s) lie
on the line <(s) = 12 .
As another example, let χ be a Dirichlet character modulo N and consider the twist
of ζ(s) by χ:
L(χ, s) =∞∑n=1
χ(n)
ns,
for <(s) > 0. When χ is not the trivial character modulo N , this function extends to an
analytic function of s ∈ C. In this case, the knowledge that L(χ, 1) is not zero implies that
there are innitely many primes congruent to a (mod N) for any a coprime to N .
1
The above examples are all instances of Artin L-functions, which are attached to Galois
representations. Indeed, a Dirichlet character modulo N can be seen as a one dimensional
representation of the Galois group of the cyclotomic eldQ(ζN ), where ζN is a primitiveNth
root of unity. As the above examples suggest, L-functions should contain some arithmetic
information at the point s = 1. This simple observation was turned into a set of deep and
relatively explicit conjectures by Stark in the seventies (in [Sta75] and the other papers in
the series). Those conjectures can be thought of as vast generalizations of the well-known
Class Number Formula, which states that
lims→1
(s− 1)ζF (s) =2r1(2π)r2hF Reg(F )
wF√|DF |
,
where r1, 2r2, hF ,Reg(F ), wF and DF are, respectively, the number of real embeddings,
the number of complex embeddings, the class number, the regulator, the number of roots
of unity and the discriminant of the number eld F . They are known to be true in two
important cases: the case of Galois representations of nite abelian extensions of Q or of an
imaginary quadratic eld K. The point that the two base elds Q and K have in common
is that one has an explicit way of generating abelian extensions of these (using cyclotomic
units for Q, and Siegel units or elliptic units for K).
A remark of Stark
Throughout this thesis, let K be an imaginary quadratic eld of discriminant D < 0.
Let F/K be an abelian extension and let ψ : Gal(F/K) −→ C× be a one-dimensional
Galois representation. Then, by class eld theory, the eld F is contained in the ray class
eld mod f for some fractional ideal f of K. It follows that ψ can be seen as a ray class
character mod f (or Hecke character mod f). The Artin L-function of this representation
is then the same as the Hecke L-function attached to ψ. By the work of Deligne-Serre, this
L-function (or, more precisely, the L-function of the induced representation from K to Q)
2
is also known to be modular. The modular form which corresponds to this representation
is a theta series.
In [Sta75], Stark noticed a connection between the special values of these L-functions
at s = 1 and the Petersson norm of the corresponding theta series when K = Q(√−23) (a
number eld of class number 3). Let ψ be one of the two non-trivial character of the class
group of K. Then ψ can be seen as a one dimensional representation of the Galois group
of H/K, where H is the Hilbert class eld of K. There is a unique newform
θψ ∈ S1(Γ0(23), χ−23),
where χ−23 is the Kronecker symbol, such that
L(θψ, s) = L(ψ, s).
As Stark noticed, the Petersson norm of θψ is
〈θψ, θψ〉 =3√
23
2πL(IndQ
K ψ, 1) = 3 log ε,
where IndQK ψ is the representation of Gal(H/Q) induced from ψ and ε is the real root of
x3 − x− 1,
which generates the Hilbert class eld of K. In this thesis, we study the Petersson inner
product of a collection of theta series which generalize the ones considered by Stark. In
particular, we see if and how Stark's observation generalizes to other imaginary quadratic
elds.
Content and structure of the thesis
This thesis is divided in three parts. In the rst one, we nd explicit formulas for the
Petersson inner product of some theta series. In the second one, we use those formulas to
3
show that the Petersson inner product of theta series can be p-adically interpolated when
p splits in K. In the last one, we treat the computational and experimental aspects of the
project.
Part I: complex formulas
Let ψ be a Hecke character of conductor OK as above which is not a genus character
(i.e. its order does not divide 2). In Chapter 4, we prove that
〈θψ, θψ〉 =−hK3w2
K
∑A∈ClK
ψ2(A) logN(A)6|∆(A)|,
where ∆ is the weight 12 and level 1 newform. As we discuss in Chapter 5, this is a direct
generalization of what Stark noticed.
More generally, one could consider higher weight theta series attached to (type A0)
Hecke characters ψ of conductor OK and innity type (2`, 0) for some ` > 0. Letting θψ be
such a theta series, we prove in Chapter 4 that
〈θψ, θψ〉 = (|D|/4)`4hKw2K
∑A∈ClK
ψ2(A)δ2`−1E2(A).
Using those formulas, one can then compute the Petersson inner product of the theta series
attached to a fractional ideals of K and ` > 0:
〈θa,`, θb,`〉 = 4(|D|/4)`∑
abc2=λcOK
λ2`c δ
2`−1E2(c),
where the sum is over all ideal classes representatives c such that abc2 is principal and
generated by λc ∈ K×. This formula follows directly from the simple relation between the
two sets of theta series.
To x notation and terminology, the classical theory of modular forms over C is in-
troduced in Chapter 1. In particular, the many Eisenstein series that come into play are
introduced. Chapter 2 is devoted to the task of computing the Petersson inner product in
4
general using the Rankin-Selberg method. In Chapter 3, the relevant results of the theory of
Complex Multiplication are introduced. The last two chapters of the rst part are devoted
to nding the explicit formulas above and discussing the weight one case in more detail.
Part II: p-adic interpolation
For ` = 0, it is well-known that one can attach weight one theta series to fractional
ideals of K. Since those theta series are not cuspidal, it makes no sense a priori to compute
their Petersson inner product. However, one could still try to use the formula for the
Petersson norm of θψ when ` = 0 to compute it. This gives formally :
〈θa,0, θb,0〉 =−1
3
∑abc2=λcOK
logN(c)6|∆(c)|.
In the second part, we show that the formulas for 〈θa,`, θb,`〉 when ` > 0 can be given
sense p-adically and that, when p splits in K, there exists a p-adic analytic function on
weight space
F :W −→ Cp
which p-adically interpolates those values. By evaluating this function at 0, which lies
outside the range of interpolation, one obtains a p-adic analogue of the above expression
for 〈θa,0, θb,0〉. This formula can be seen as a p-adic analogue of Kronecker's First Limit
formula.
The techniques used to p-adically interpolate the above quantities use the same techni-
ques that Katz used to construct the p-adic L-function attached to Hecke characters of
imaginary quadratic elds in which p splits (see [Kat76]). Those techniques rely on the
theory of p-adic modular forms. The denition of these objects and the transition from
the classical theory modular forms over C to the theory of algebraic and eventually p-adic
modular forms is done in Chapter 6. We then re-visit the theory of complex multiplication
5
from an algebraic point of view in Chapter 7. Finally, the theory introduced in those two
chapters is used to p-adically interpolate the Petersson inner product of theta series.
Part III: computations
A signicant part of my research time was devoted to doing computations and nding
algorithms to compute the Petersson inner product of theta series, and so I felt it was
relevant to include some computations in this thesis. The computations are related to some
of the sections of this thesis. The sections which are marked with the symbol F have
computations attached to them (for example, Section 2.3 is one of them). The last chapter
of this thesis, Chapter 13, is exploratory and contains a series of experiments which lead to
various observations, conjectures and open questions.
To maximize the fun, the reader should read this last part with a computer nearby to
reproduce the computations in PARI/GP and run his or her own experiments!
6
Part I
Complex formulas
7
CHAPTER 1Modular forms over C
In this rst chapter, we give an overview of the classical theory of modular forms over
C, while in the next part we give an overview of the algebraic and p-adic theories. This is
done mostly to x notation. Indeed, the notation is not standard (at least for Eisenstein
series and Poincarré series) and we need results from a few references (namely [DS05],
[RBvdG+08], [Shi10], [MM06] and [Coh07]).
We assume that the reader is familiar with the basic theory of classical modular forms.
If this is not the case, good references are [RBvdG+08, Part 1] and [DS05].
1.1 The spaces GL+, H, LΓ and Y (Γ)
As in [Kat76, Sec.1.0], let
GL+ = (ω1;ω2) ∈ C2 : =(ω1/ω2) > 0 (1.1)
be the set of positively oriented R-bases of C. The notation (a; b) stands for a 2×1 column
vector (the semi-colon indicates a change of row, instead of column). This set is naturally
equipped with a structure of a smooth manifold by inclusion into C2.
As usual, let SL2(Z) denote the group of 2 × 2 matrices with integer entries and
determinant 1. This group acts on GL+ on the left asa b
c d
ω1
ω2
=
aω1 + bω2
cω2 + dω2
. (1.2)
8
The group C× also acts on GL+ by scaling. Finally, GL+ is equipped with an involution ρ
dened as
(ω1;ω2)ρ = (ω1;−ω2), (1.3)
where the bar denotes complex conjugation. The involution ρ is essentially complex conju-
gation, just slightly modied to preserve the orientation of bases.
The map
(ω1;ω2) 7→ ω1/ω2 (1.4)
from GL+ to the complex upper-half plane
H = τ ∈ C : =(τ) > 0
is surjective and induces a bijection
GL+/C× ' H. (1.5)
Note that this map also has a natural section
τ 7→ (τ ; 1) : H −→ GL+/C×. (1.6)
The action of SL2(Z) on GL+ descends to the quotient by C× and the induced action on
H via the above bijection corresponds to the usual action of SL2(Z) on H:a b
c d
τ =aτ + b
cτ + d. (1.7)
The involution ρ on H sends τ to −τ .
9
The quotient of GL+ by SL2(Z) is the set of lattices in C, denoted L. More generally,
let N be a strictly positive integer and let
Γ(N) =
a b
c d
∈ SL2(Z) : a, d ≡ 1 (mod N) and b, c ≡ 0 (mod N)
(1.8)
be the kernel of the reduction mod N map on SL2(Z). A congruence subgroup is a sub-
group Γ of SL2(Z) which contains the subgroup Γ(N) for some N . The two most common
examples of congruence subgroups are
Γ0(N) =
a b
c d
∈ SL2(Z) : c ≡ 0 (mod N)
(1.9)
and
Γ1(N) =
a b
c d
∈ SL2(Z) : a, d ≡ 1 (mod N) and c ≡ 0 (mod N)
. (1.10)
For a general congruence subgroup Γ, dene
LΓ = Γ \GL+. (1.11)
The group C× and the involution ρ act on LΓ in the obvious way. Note that ρ sends a
lattice L ∈ LΓ to L, as one might expect.
The quotient of GL+ by the action of a congruence subgroup Γ and C× is denoted
Y (Γ) and is called the open modular curve for Γ. It has the structure of a (non-compact)
Riemann surface (see [DS05, Ch.2]).
10
To summarize, the four spaces introduced in this section t into the diagram
GL+
""||
H
!!
LΓ
||
Y (Γ)
(1.12)
where all the arrows are surjective and compatible with the actions of SL2(Z), C× and ρ
on the various spaces.
1.2 Weight and q-expansions
Let (k, s) ∈ Z×C. As in [Kat76, Sec.1.1], a C∞ function F : GL+ −→ C is said to be
of weight (k, s) if
F (λ(ω1;ω2)) = λ−k|λ|−2sF ((ω1;ω2)) for all λ ∈ C×. (1.13)
The notion of weight also applies to C∞ functions f : LΓ −→ C, since they can be viewed
as Γ-invariant functions on GL+. Such functions on LΓ are usually called homogeneous of
weight (k, s). A C∞ function f : H −→ C is said to be of weight (k, s) for Γ if
f
(aτ + b
cτ + d
)= (cτ + d)k|cτ + d|2sf(τ) for all
a b
c d
∈ Γ. (1.14)
For simplicity, a function of weight (k, 0) is said to be of weight k.
Now let F be a C∞ homogeneous function of weight (k, s) on LΓ. Then F induces a
C∞ function f on H which is of weight (k, s) for Γ by letting
f(τ) = F ((τ ; 1)). (1.15)
11
Conversely, any C∞ function f on H of weight (k, s) for Γ induces a C∞ function F on LΓ
of weight (k, s) by dening
F ([ω1, ω2]) = ω−k2 |ω2|−2sf(ω1/ω2). (1.16)
Here, [ω1, ω2] denotes the Γ-equivalence class of (ω1;ω2) ∈ GL+ (if Γ = SL2(Z), then
[ω1, ω2] corresponds to the lattice spanned by ω1 and ω2 in C). One can check that those
maps induce a bijection between the set of C∞ homogeneous functions of weight (k, s) on
LΓ and the set of C∞ functions of weight (k, s) for Γ on H.
Now let f : H −→ C be a holomorphic function of weight k for Γ ⊃ Γ(N). Then
f(τ +N) = f(τ),
since 1 N
0 1
∈ Γ(N) ⊆ Γ.
It follows that f(τ) can be expressed as a holomorphic function of q = e2πiτ/N on the open
unit disc in C with the origin removed. This function of q, still denoted f , has a Laurent
expansion
f(q) =∑n
an(f)qn (1.17)
around the origin. This expression is called the q-expansion (or sometimes Fourier expan-
sion) of f .1 The complex numbers an(f) are called the Fourier coecients of f . Using
the above correspondence, one can also dene the q-expansion of a holomorphic function
of weight k on LΓ.
1 Note that this denition diers from the one given in [Kat76, Sec.1.1], but agrees withthe one in [DS05]. The convention adopted here is more common these days.
12
1.3 Modular forms
In the rst part of this thesis, modular forms over C will be viewed in two equivalent
ways.
Modular forms as functions on the upper half-plane
A weakly holomorphic modular form of weight k for a congruence subgroup Γ is a
holomorphic function f : H −→ C of weight k for Γ whose translates by elements of SL2(Z)
have meromorphic q-expansions (i.e. f is holomorphic onH and meromorphic at the cusps).
Note that this denition makes sense since the translate of f by an element γ ∈ SL2(Z) is
holomorphic and of weight k for the congruence subgroup γ−1Γγ.
If the q-expansions of the SL2(Z)-translates of f are holomorphic, we call f a (holo-
morphic) modular form of weight k for Γ. If those q-expansions all have no constant term,
we call f a cusp form. The C-vector spaces of modular forms (resp. cusp forms) of weight
k for Γ are denoted
Mk(Γ) (resp. Sk(Γ)).
If f is just C∞ (i.e. not necessarily holomorphic on H) and of weight (k, s) for Γ, we
call f a C∞ (or non-holomorphic) modular form of weight (k, s) for Γ.
Modular forms as functions on LΓ
Using the correspondence of the previous section, one can equivalently dene modular
forms as weight k functions on LΓ satisfying certain analyticity conditions. In particular,
modular forms for SL2(Z) can be viewed as functions on lattices.
Some important examples
A rst simple example of a C∞ modular form of weight (0,−1) for SL2(Z) is
a(L) = area(C/L).
13
As a function on GL+, it is given by
a((ω1;ω2)) = =(ω1ω2) =ω1ω2 − ω1ω2
2i.
As a function on H, it is simply
a(τ) = a((τ ; 1)) = =(τ).
A second less trivial example is given by the ∆ function, which is easier to dene via
its q-expansion as
∆(q) = q
∞∏n=1
(1− qn)24. (1.18)
This holomorphic function on H is a weight 12 cusp form for SL2(Z) (see [RBvdG+08,
Prop.7]).
An example of weakly holomorphic modular form is given by 1/∆, since ∆ does not
vanish on H.
The Eisenstein series, introduced in the next section, give a lot of examples of modular
forms of various weight on SL2(Z). In particular, they can be used to construct other
modular forms. For example,
1728∆ = (240E4)3 − (504E6)2, (1.19)
where E4 and E6 are the weight 4 and weight 6 Eisenstein series, respectively. They also
enter in the construction of the most important modular function on SL2(Z), the j-invariant,
which is dened as
j(q) =(240E4(q))3
∆(q)
= q−1 + 744 + 196884q + 21493760q2 +O(q3)
(see [RBvdG+08, Sec.2.4]).
14
The q-expansion principle
The q-expansion principle says that modular forms are determined by their q-expansions.
This is obvious for modular forms over C, since the can be viewed as function on H and q
is given by e2πiτ for τ ∈ H. However, for algebraic modular forms, dened in the second
part, this is a less trivial and very useful fact.
1.4 Holomorphic and C∞ Eisenstein series
Eisenstein series are very useful in the theory of modular forms in general and they
play a central role in this thesis. For the rest of this section, let k ≥ 0 be an integer. The
simplest example of an Eisenstein series is
Gk(L) =∑λ∈L
′ 1
λkfor k > 2, (1.20)
where L ∈ L is a lattice and the ′ symbol means that the term corresponding to λ = 0 is
excluded. Then Gk is a weight k modular form on SL2(Z). To prove this, it is preferable
to rst write Gk as a function on H
Gk(τ) =∑m,n∈Z
′ 1
(mτ + n)kfor k > 2, (1.21)
where the ′ symbol means that the term (m,n) = (0, 0) is excluded. For k > 2 this sum
converges absolutely. This proves that Gk(τ) is holomorphic on H and, by rearranging the
sum, that it is of weight k for SL2(Z). Finally, a direct application of Lipschitz's formula
shows that Gk has q-expansion
Gk(q) = 2ζ(k) + 2(2πi)k
Γ(k)
∞∑n=1
σk−1(n)qn, (1.22)
where σk−1(n) =∑
d|n dk−1 and ζ(s) is the Riemann zeta function (see [RBvdG+08,
Prop.5], keeping in mind that our notation dier). It is convenient to normalize Gk in
15
such a way that its q-expansion has rational coecients. Dene
Ek =1
2
Γ(k)
(2πi)kGk, (1.23)
so that
Ek(q) = −Bk2k
+∞∑n=1
σk−1(n)qn, (1.24)
where Bk is the k Bernoulli number (note that Euler's evaluation of ζ(k) for k > 0 even in
terms of Bernoulli numbers is used). For example,
E4(q) =1
240+ q + 9q2 + 28q3 + . . . (1.25)
and
E6(q) = − 1
504+ q + 33q2 + 244q3 + . . . . (1.26)
Those Eisenstein series can be generalized in many ways. In this text, the following
three generalizations will be needed:
1. Eisenstein series of weight k = 2;
2. C∞ Eisenstein series of weight (k, s);
3. Eisenstein series of higher level.
C∞ Eisenstein series for SL2(Z)
When k = 2, the series dening Gk in (1.21) diverges. One way to dene a weight 2
Eisenstein series is to introduce a parameter s ∈ C in the summation dening Gk:
Gk,s(L) =∑λ∈L
′ 1
λk|λ|2sfor <(2s) + k > 2. (1.27)
Then Gk,s is a C∞ modular form of weight (k, s) for SL2(Z). As above, this can be proven
by writing Gk,s(L) as an absolutely convergent sum over the upper half-plane
Gk,s(τ) =∑m,n∈Z
′ 1
(mτ + n)k|mτ + n|2sfor <(2s) + k > 2. (1.28)
16
When k > 2, one can let s be 0 in (1.27) or (1.28) and recover the Eisenstein series
dened above:
Gk,0 = Gk for k > 2. (1.29)
When k = 2, the modular form G2,s is not dened at s = 0. However, one can dene G2(τ)
as the limit of G2,ε(τ) as ε tends to 0 from the right. Then one can prove that
G2(τ) = −8π2
(−1
24+∞∑n=1
σ1(n)qn
)− π
=(τ)(1.30)
(see [RBvdG+08, Prop.6], keeping in mind again that our notation dier). This function
G2 is a C∞ weight 2 modular form for SL2(Z). Using the normalization introduced in
(1.23), one is lead to dene
E2(τ) =1
2
1
(2πi)2G2(τ), (1.31)
which has "q-expansion"1
8π=(τ)− 1
24+∞∑n=1
σ1(n)qn. (1.32)
Note that since E2 is a modular form for SL2(Z), it can be evaluated on L using the
correspondence (1.16).
The C∞ Eisenstein series Gk,s can also be completed by introducing some factors at
innity as follows:
G∗k,s(L) =1
2Γ(s+ k)
(a(L)
π
)sGk,s(L) for <(2s) + k > 2. (1.33)
Equivalently,
G∗k,s(τ) =1
2Γ(s+ k)
(=(τ)
π
)sGk,s(τ) for <(2s) + k > 2. (1.34)
17
Then G∗k,s is a C∞ modular form of weight k (recall that a(L) and =(τ) have weight (0,−1))
and
Ek = (2πi)−kG∗k,0 for k ≥ 2. (1.35)
Moreover, the following Theorem holds.
Theorem 1. For every k ≥ 0 and τ ∈ H, the C∞ modular form G∗k,s(τ) can be continued
to a meromorphic function of s ∈ C, which is entire if k > 0, and has poles with residues −12
and 12 at s = 0 and s = 1, respectively, if k = 0. Moreover, G∗k,s(τ) satises the following
functional equation:
G∗k,s(τ) = G∗k,1−k−s(τ).
Proof. See [Shi10, Theorem 9.7]. Note that
G∗k,s(τ) =1
2Z1k(τ, s; 0, 0)
in Shimura's notation.
When k = 0, Kronecker's rst limit formula gives the constant term of the Taylor
expansion of G∗0,s at s = 1.
Theorem 2 (Kronecker's First Limit Formula). Around s = 1,
G∗0,s(L) =1/2
s− 1+ (γ − log 2)− 1
12log(a(L)6|∆(L)|) +O(s− 1),
where γ is Euler's constant.
Proof. This is proved in [Coh07, Thm.10.4.6], for example.
18
C∞ Eisenstein series on Γ0(N)
Let N ≥ 1 be an integer. One can dene a C∞ Eisenstein series for Γ0(N) as
Gk,s,N (τ) =∑
t∈(Z/NZ)×
∑(m,n)≡(0,t) (mod N)
′ 1
(mτ + n)k|mτ + n|2s
=∑m,n∈Z
′ 1N (n)
(mNτ + n)k|mNτ + n|2sfor <(2s) + k > 2,
where 1N (n) = 1 if gcd(n,N) = 1 and 0 otherwise (this is the trivial character mod N).
A straightforward computation shows that Gk,s,N is a C∞ weight (k, s) modular form
for Γ0(N). Dene
G∗k,s,N (τ) =1
2Γ(k + s)
(=(τ)
π
)sGk,s,N (τ) for <(2s) + k > 2. (1.36)
Then one has the following
Theorem 3. For k ≥ 0 and N > 1, the C∞ modular form G∗k,s,N can be continued to a
meromorphic function of s ∈ C, which is entire if k > 0, and has a pole with residue ϕ(N)2N2
at s = 1 if k = 0. Here, ϕ(N) is Euler's phi function.
Proof. This is [Shi10, Theorem 9.7] again, using the fact that
G∗k,s,N (τ) =1
2
∑t∈(Z/NZ)×
ZNk (τ, s; 0, t)
in Shimura's notation.
Recall that ϕ(N) has the simple expression
ϕ(N) = N∏p|N
(1− p−1). (1.37)
19
We conclude this section by introducing Poincaré series, which are used in the Rankin-
Selberg method. For γ =
a b
c d
∈ SL2(Z), dene the automorphy factor j(γ, τ) as
j(γ, τ) = cτ + d. (1.38)
Then the Poincaré series Pk,s,N is dened as
Pk,s,N (τ) = =(τ)s∑
γ∈Γ∞\Γ0(N)
j(γ, τ)−k|j(γ, τ)|−2s for <(2s) + k > 2, (1.39)
where
Γ∞ =
±1 m
0 1
: m ∈ Z
.
Taking the set a b
cN d
: ad− bcN = 1, d > 0
as a set of representatives for the quotient Γ∞ \ Γ0(N) (see [MM06, Lemma 7.1.6]), we see
that
Pk,s,N (τ) = =(τ)s∑
gcd(cN,d)=1,d>0
1
(cNτ + d)k|cNτ + d|2s.
It follows that1
2=(τ)sGk,s,N (τ) = ζN (2s+ k)Pk,s,N (τ), (1.40)
where
ζN (s) =∏p|N
(1− p−s)ζ(s).
When k = 0, one has
P0,s,N (τ) =∑
γ∈Γ∞\Γ0(N)
=(γτ)s for <(s) > 1, (1.41)
20
since
=(γτ) = |j(γ, τ)|−2=(τ) (1.42)
(note that this equation is just saying that =(τ) has weight (0,−1)). Using relation (1.40),
one sees that P0,s,N extends to a meromorphic function of s ∈ C. Completing G0,s,N in
(1.40) and taking residues at s = 1, one sees using Theorem 3 when N > 1, Theorem 1
when N = 1 and (1.37) that
1
2N
∏p|N
(1− p−1) =ϕ(N)
2N2= ress=1G
∗0,s,N (τ) =
ζN (2)
πress=1P0,s,N (τ).
It follows that
ress=1P0,s,N (τ) =3
Nπ
∏p|N
(1 + p−1)−1 = Vol(Γ0(N) \ H)−1, (1.43)
where the volume is computed in the hyperbolic measure on H (see [RBvdG+08, Sec.1.3]).
1.5 Operators on Mk(Γ0(N), χ)
A detailed exposition of the material of this section can be found in [DS05, Ch.5], for
example.
A basic, but non-trivial property of the spaces Mk(Γ1(N)) is that they are nite
dimensional over C (see [RBvdG+08, Prop.3]). For Γ = SL2(Z), one can show that the
Eisenstein series E4 and E6 introduced above generate the graded C-algebra of modular
forms for SL2(Z) (see [RBvdG+08, Prop.4]):
⊕k≥0
Mk(SL2(Z)) = C[E4, E6]. (1.44)
It follows that,
Mk(SL2(Z)) = C− span of Ea4Eb6 : 4a+ 6b = k,
21
which shows in particular that Mk(SL2(Z)) is nite dimensional for every k ≥ 0. For
example,
1. M0(SL2(Z)) = C;
2. M2(SL2(Z)) = 0;
3. Mk(SL2(Z)) = CE4,CE6,CE24 ,CE4E6 for k = 4, 6, 8 and 10, respectively;
4. M12(SL2(Z)) = CE34 ⊕ CE2
6 and S12(SL2(Z)) = C∆.
Throughout this section, the q-expansion of a modular form f ∈Mk(Γ1(N)) is denoted
f(q) =
∞∑n=0
anqn.
The diamond operators and the space Mk(Γ0(N), χ)
Let f be a modular form of weight k on Γ1(N). The homomorphism
Γ0(N) −→ (Z/NZ)× (1.45)
which sends a matrix γ ∈ Γ0(N) to its lower right entry is surjective and has Γ1(N) as
kernel. It follows that Γ0(N) normalizes Γ1(N), and so Γ0(N) acts on Mk(Γ1(N)) by
precomposition (i.e. (fγ)(τ) = f(γτ) for γ ∈ Γ0(N)). Since Γ1(N) acts trivially on that
space by denition, one gets an induced action of
Γ0(N)/Γ1(N) ∼= (Z/NZ)× (1.46)
on that space. To each d ∈ (Z/NZ) one attaches a linear operator 〈d〉, called a diamond
operator, dened as 0 if gcd(d,N) > 1 and as
(〈d〉f)(τ) = f(γdτ), (1.47)
where γd is any preimage of d (mod N) under the isomorphism (1.46), otherwise.
22
Now let
χ : (Z/NZ)× −→ C×
be any Dirichlet character, extended to Z/NZ in the usual way and dene
Mk(Γ0(N), χ) = f ∈Mk(Γ1(N)) : 〈d〉f = χ(d)f, ∀d ∈ Z/NZ. (1.48)
An element of Mk(Γ0(N), χ) is called a modular form of weight k, level N and character
χ. By denition, if f is such a modular form, it satises the functional equation
f
(aτ + b
cτ + d
)= χ(d)(cτ + d)kf(τ) for all
a b
c d
∈ Γ0(N). (1.49)
The space Sk(Γ0(N), χ) is dened in a similar way (just note that the diamond operators
preserve Sk(Γ1(N))).
The Up operator on Mk(Γ0(N), χ)
Let p be a prime number and let f ∈ Mk(Γ0(N), χ). The Up operator is dened on
q-expansions as
(Upf)(q) =∞∑n=0
apnqn. (1.50)
Viewing f as a function on the upper half-plane, one can express Up as
(Upf)(τ) =1
p
p∑j=1
f
(τ + j
p
). (1.51)
If p|N , the operator Up preserves the spaces Mk(Γ0(N), χ) and Sk(Γ0(N), χ) (see
[DS05, Prop.5.2.1]). However, if p - N , it does not.
23
The Vm operator on Mk(Γ0(N), χ)
Let m be an integer and f ∈ Mk(Γ0(N), χ) be as above. The Vm operator is dened
on q-expansions as
(Vmf)(q) =∞∑n=0
anqmn. (1.52)
Equivalently, Vm has the simple expression
(Vmf)(τ) = f(mτ). (1.53)
The operator Vm never preserves the space Mk(Γ0(N), χ); it sends f to a form of level mN
(this is exercise 1.2.11 in [DS05]).
As operators on q-expansions, the Vm operators satisfy the following relations:
Vm =∏pr‖m
(Vp)r.
The Hecke operator Tm on Mk(Γ0(N), χ)
First, let p be a prime number and let f be as above. The Hecke operator Tp of weight
k on Mk(Γ0(N), χ) is dened as
Tp = Up + χ(p)pk−1Vp. (1.54)
Here the character χ is extended to all Z/NZ in the usual way, so that
Tp = Up if p|N.
The operator T1 is dened as the identity and Tpr , for r ≥ 2, is dened recursively as
Tpr = TpTpr−1 − χ(p)pk−1Tpr−2 . (1.55)
24
Note that this equation simplies to
Tpr = (Up)r
if p|N . Finally, for positive integers m, dene
Tm =∏pr‖m
Tpr . (1.56)
The denition of Tm can be written more succinctly as an equality between formal power
series in X as
∞∑m=1
TmXm =
∏p|N
(1− TpX)−1∏p-N
(1− TpX + χ(p)pk−1X2)−1. (1.57)
The Hecke operators have the following explicit expression on q-expansions:
(Tmf)(q) =∞∑n=0
∑d|(m,n)
χ(d)dk−1amn/d2
qn, (1.58)
where the inner sum is taken over positive common divisors of m and n (see [DS05,
Prop.5.3.1]).
An important fact is that the Hecke operators Tm of weight k preserve the spaces
Mk(Γ0(N), χ) and Sk(Γ0(N), χ) (see [DS05, Prop.5.2.2]), and that they all commute with
each other (see [DS05, Prop.5.2.4]).
The Hecke operators also have nice expressions on modular forms considered as functi-
ons on LΓ. For example, in level one (i.e. Γ = SL2(Z)) one has
(Tmf)(L) = mk−1∑
L′⊆L:[L:L′]=m
f(L′), (1.59)
where the sum is taken over all sub-lattices of index m in L (see [RBvdG+08, Sec.4.1]).
25
The ρ operator on Mk(Γ0(N), χ)
It was shown in Section 1.1 that the spaces GL+, LΓ and H could be equipped with
an action of complex conjugation, which was denoted ρ. One can also dene an action of ρ
on f ∈Mk(Γ0(N), χ) by letting
fρ(q) =
∞∑n=0
anqn. (1.60)
Then fρ is a modular form in Mk(Γ0(N), χ), where as usual χ(n) = χ(n) = χ(n)−1 for any
n prime to the level (this is in [Shi76, Sec.2], for example). Equivalently,
fρ(τ) = f(τρ) = f(−τ), (1.61)
when τ ∈ H or
fρ(L) = f(Lρ), (1.62)
when L ∈ LΓ1(N).
Using (1.24), it is clear that
Eρk = Ek for k > 2. (1.63)
In weight 2, it follows from (1.32) that
E2(−τ) = E2(τ). (1.64)
It follows that
Ek(•ρ) = Ek(•) for k ≥ 2, (1.65)
where • ∈ L or H.
26
1.6 Petersson inner product and Atkin-Lehner Theory
The Petersson inner product
Let x and y be the standard coordinates on the upper half-plane, so that
τ = x+ iy,
and let
dµ =dxdy
y2
be the SL2(Z)-invariant hyperbolic measure onH. For any two cusp forms f, g ∈ Sk(Γ0(N), χ),
the function
F (τ) = f(τ)g(τ)=(τ)k
is a C∞ modular form of weight 0 for Γ0(N). Since f(τ) goes to 0 exponentially fast as τ
approaches the cusps, so does F (τ). The Petersson inner product of f and g is dened as
〈f, g〉 =
∫∫Γ0(N)\H
f(τ)g(τ)=(τ)kdµ, (1.66)
which is a well-dened integral by the properties of F (see [DS05, Sec.5.4]). This denes a
Hermitian inner product on Sk(Γ0(N), χ).
The adjoint of the Hecke operator Tm with respect to the Petersson inner product is,
by denition, the operator T ∗m on Sk(Γ0(N), χ) such that
〈Tmf, g〉 = 〈f, T ∗mg〉.
Proposition 1. If p - N , the adjoint of Tp under the Petersson inner product is
T ∗p = χ(p)Tp.
It follows that the Hecke operators Tm for m prime to N are normal (i.e. they commute
with their adjoint).
27
Proof. [DS05, Theorem 5.5.3].
Newforms
Since the Hecke operators Tm, for gcd(m,N) = 1, form a commuting family of normal
operators on the space Sk(Γ0(N), χ), this space has a basis of orthogonal eigenvectors for
all Tm with gcd(m,N) = 1. Those eigenvectors are called eigenforms.
Fix an integer m prime to the level, let f be an eigenform and let λm be such that
Tmf = λmf.
Then
a1(Tmf) = λma1(f).
Using (1.58), one sees that
a1(Tmf) = am(f).
It follows that
am(f) = a1(f)λm, (1.67)
which implies that the Fourier coecients with index prime to the level of an eigenform are
determined by its eigenvalues and the leading term of its q-expansion.
Dene the space of oldforms Soldk (Γ0(N), χ) as the C-vector-space spanned by the
modular forms Vdg for d|N/M , d > 1, M |N and g ∈ Sk(Γ0(M), χ), where the modular
form Vdg of level Md is viewed as a modular form of level N under the natural inclusion
Sk(Γ0(dM), χ) ⊆ Sk(Γ0(N), χ).
The new subspace of Sk(Γ0(N), χ), denoted Snewk (Γ0(N), χ), is dened as the orthogonal
complement (under the Petersson inner product) of Soldk (Γ0(N), χ).
28
Proposition 2. The subspaces Soldk (Γ0(N), χ) and Snew
k (Γ0(N), χ) are stable under the
Hecke operators.
Proof. [DS05, Proposition 5.6.2].
If f is in the old subspace, it is clear that an(f) = 0 whenever gcd(n,N) = 1. The
Main Lemma in Atkin-Lehner theory says that the converse also holds.
Theorem 4 (Main Lemma). If f ∈ Sk(Γ0(N), χ) has Fourier expansion f(q) =∑∞
n=1 an(f)qn
with an(f) = 0 whenever gcd(n,N) = 1, then f ∈ Soldk (Γ0(N), χ).
Proof. [DS05, Theorem 5.7.1].
If f is in an eigenform in the new subspace, a1(f) 6= 0 by the Main Lemma and (1.67).
Therefore it makes sense to dene a newform as an eigenform which is normalized in such
a way that a1(f) = 1. Using the Main Lemma and (1.67) again, one can then prove the
following
Proposition 3. Let f be a newform. Then f is an eigenvector for all Hecke operators with
eigenvalue
Tmf = am(f)f.
In particular, all the Fourier coecients of a newform, hence the newform itself, are
determined by the eigenvalues of that newform under the action of the Hecke operators.
1.7 Dierential operators on modular forms
Dene the D operator as
D =1
2πi
∂
∂τ= q
d
dq, (1.68)
so that
D
( ∞∑n=0
anqn
)=
∞∑n=1
nanqn (1.69)
29
on q-expansions. If f ∈ Mk(Γ1(N)) is a modular form, Df is holomorphic, but does not
behave well under the action of Γ1(N). In fact,
Df
(aτ + b
cτ + d
)= (cτ + d)k+2Df(τ) +
k
2πic(cτ + d)k+1f(τ) (1.70)
(see [RBvdG+08, Eq.52]). Note that if there was no second term, Df would be a modular
form of weight k+2. To get rid of the second term, one can introduce a "correction factor".
This leads to the Shimura-Maass operator
δk = D − k
4π=(τ)=
1
2πi
(∂
∂τ+
k
τ − τ
). (1.71)
Using (1.42), one can show that δkf has weight k + 2 under Γ1(N) (see [RBvdG+08, after
Eq.55]). However, it is not holomorphic anymore (but still C∞). Note that the D operator
extends to C∞ functions by letting
∂
∂τ=
1
2
(∂
∂x− i ∂
∂y
). (1.72)
One can therefore iterate this operator. Dene
δrk = δk+2r−2 . . . δk+2 δk.
The Shimura-Maass operator on modular forms on SL2(Z)
The Shimura-Maass operator on Mk(SL2(Z)) can expressed in terms of the Eisenstein
series E2, E4 and E6 as follows:
δ2E2 =5
6E4 − 2E2
2 (1.73)
δ4E4 =7
10E6 − 8E2E4 (1.74)
δ6E6 =400
7E2
4 − 12E2E6. (1.75)
30
To prove the second equation, for example, rst compute
(δ4E4 +8E2E4)(τ) =
(DE4(τ)− E4(τ)
π=(τ)
)+
(E4(τ)
π=(τ)− (PE4)(τ)
3
)= DE4(τ)− (PE4)(τ)
3,
where
P (τ) = E2(τ)− 1
8π=(τ)= − 1
24+∞∑n=1
σ1(n)qn.
Since the left hand side is of weight 6 for SL2(Z) and the right hand side is holomorphic,
δ4E4 + 8E2E4 is a modular form of weight 6 for SL2(Z), so it is a multiple of E6.
It follows from the above formulas that the graded C-algebra generated by E2, E4 and
E6 is preserved by the Shimura-Maass operator (by acting on each graded piece). A weight
k element of this algebra is called a nearly holomorphic modular form of weight k on SL2(Z).
The Shimura-Maass operator on C∞ modular forms on SL2(Z)
Recall that the C∞ Eisenstein series G∗k,s has weight k. Therefore is makes sense to
apply the Shimura-Maass operator to it. Using term-by-term dierentiation, one shows
that
δkG∗k,s = (2πi)−2G∗k+2,s−1 (1.76)
for <(s) large enough and then for all s (see [Shi10, Section 9.4]).
1.8 The L-function of modular forms
Let f =∑∞
n=0 anqn ∈ Mk(Γ0(N), χ) be a modular form. One can naturally attach a
Dirichlet series to it:
L(f, s) =
∞∑n=1
anns. (1.77)
This function of s ∈ C converges for <(s) > k/2 + 1 if f is a cusp form and for <(s) > k
otherwise. When f is a newform, it is an L-function.
Theorem 5. Let f =∑∞
n=1 anqn ∈ Sk(Γ0(N), χ) be a newform and dene
L∗(f, s) = N s/2ΓC(s)L(f, s) for <(s) > k/2 + 1, (1.78)
31
where ΓC(s) = (2π)−sΓ(s). Then
1. L∗(f, s) extends to a holomorphic function on C;
2. L∗(f, s) satises the functional equation
L∗(f, s) = εL∗(f, k − s),
where ε = ±1;
3. For <(s) > k/2 + 1, L(f, s) can be expressed as an Euler product
L(f, s) =∏p
(1− app−s + χ(p)pk−1p−2s)−1.
Proof. See [DS05, Theorem 5.10.2] for the rst two points and [DS05, Theorem 5.9.2] for
the third point.
32
CHAPTER 2Computing the Petersson inner product of cusp forms
In this chapter, the main tools and formulas to compute the Petersson inner product
of modular forms are introduced, in decreasing order of generality.
2.1 The Rankin-Selberg method
The Rankin-Selberg method is a simple but clever manipulation of integrals that leads
to an explicit relation between the Petersson inner product of two modular forms and a
residue of a Dirichlet series. Another exposition of this method is given in [Shi76, Sec.2],
for example.
Let f, g ∈ Sk(Γ0(N)) be two modular forms with q-expansions∑anq
n and∑bnq
n, re-
spectively. Recall that the Poincaré series P0,s,N of weight (0, 0) for Γ0(N) can be expressed
as
P0,s,N (τ) =∑
γ∈Γ∞\Γ0(N)
=(γτ)−s
for <(s) large enough. Since the function
F (τ) = f(τ)g(τ)=(τ)k
is also of weight (0, 0) for Γ0(N), the integral
I(s) =
∫∫Γ0(N)\H
F (τ)P0,s,N (τ)dµ(τ) (2.1)
converges for <(s) > 1. The idea of the Rankin-Selberg method is to compute the residue
at s = 1 of I(s) in two dierent ways.
33
On the one hand, since P0,s,N (τ) has a simple pole at s = 1 whose residue does not
depend on τ , it follows that
ress=1I(s) = Vol(Γ0(N) \ H)−1〈f, g〉, (2.2)
by (1.43) and the denition of the Petersson inner product of f and g.
On the other hand, for <(s) large enough∫∫Γ0(N)\H
F (τ)P0,s,N (τ)dµ(τ) =
∫∫Γ0(N)\H
∑γ∈Γ∞\Γ0(N)
F (τ)=(γτ)sdµ(τ)
=∑
γ∈Γ∞\Γ0(N)
∫∫Γ0(N)\H
F (τ)=(γτ)sdµ(τ)
=∑
γ∈Γ∞\Γ0(N)
∫∫Γ0(N)\H
F (γτ)=(γτ)sdµ(τ)
=
∫∫Γ∞\H
F (τ)=(τ)sdµ(τ).
A fundamental domain for the region Γ∞ \ H is given by τ ∈ H : 0 ≤ <(τ) ≤ 1, so that
the last integral can be written as∫∫Γ∞\H
F (τ)=(τ)sdµ(τ) =
∫ ∞0
∫ 1
0F (x+ iy)ys−2dxdy.
Now
f(τ)g(τ) =∞∑
m,n=1
anbme2πinze−2πimτ =
∞∑m,n=1
anbme2πi(n−m)xe−2π(m+n)y,
so ∫ 1
0F (x+ iy)dx =
∞∑n=1
anbne−4πnyyk
and ∫ ∞0
(∫ 1
0F (x+ iy)ysdx
)dy
y2=
Γ(s+ k − 1)
(4π)s+k−1
∞∑n=1
anbnns+k−1
.
34
Taking the residue at s = 1, we get
ress=1I(s) =Γ(k)
(4π)kress=kD(f, gρ, s), (2.3)
where
D(f, g, s) =∞∑n=1
anbnns
.
Putting (2.2) and (2.3) together, one obtains the following
Theorem 6 (Rankin-Selberg). Let f, g ∈ Sk(Γ0(N), χ). Then
〈f, g〉 = Vol(Γ0(N) \ H)Γ(k)
(4π)kress=kD(f, gρ, s),
where
Vol(Γ0(N) \ H) =π
3N∏p|N
(1 + p−1).
Thanks to this theorem, the computation of the Petersson inner product is reduced to
the computation of the residue of a Dirichlet series at s = k.
2.2 Rankin- Selberg convolution and symmetric square L-functions
Suppose now that f and g are common eigenforms for all Hecke operators with a1 =
b1 = 1. Those conditions hold when f and g are newforms, for example. Let
L(f, s) =∏p
(1− app−s + χ(p)pk−1p−2s)−1 =∏p
((1− αpp−s)(1− βpp−s))−1
be the Euler product of L(f, s). Recall that χ(p) = 0 if p|N . By convention, let βp = 0 if
p|N . It follows that
αp + βp = ap (2.4)
and
αpβp =
χ(p)pk−1 if p - N
0 if p|N.
35
The numbers αp and βp are called the roots of the Hecke polynomial of f at p. Similarly,
let α′p and β′p be the roots of g at p.
Twisted modular forms
Let ω be a primitive Dirichlet character mod M and let χ0 be the primitive Dirichlet
character corresponding to χ. Then the q-expansion
∞∑n=1
ω(n)anqn (2.5)
is the q-expansion of a modular form of level lcm(N,M2,M cond(χ0)), character χ0ω2 and
weight k (see [Shi71][Prop 3.64]). This modular form is called the twist of f by ω and is
denoted fω. Here χ0ω2 is viewed as a lcm(N,M2,M cond(χ0))-periodic function on Z (in
particular, it may not be a primitive character).
It is not hard to see that the roots of fω at p are ω(p)αp and ω(p)βp.
Twisted symmetric square L-function
The symmetric square L-function of f twisted by ω is dened by the Euler product
L(Sym2 f, ω, s) =∏p
[(1− ω(p)α2pp−s)(1− ω(p)αpβpp
−s)(1− ω(p)β2pp−s)]−1 (2.6)
for <(s) large enough.
The following theorem of Shimura is very useful when dealing with L(Sym2 f, ω, s).
Theorem 7. Let f ∈ Mk(Γ0(N), χ) be a common eigenfunction for all Hecke operators
and let L(Sym2 f, ω, s) be the symmetric square L-function of f twisted by ω. Then the
function
R(Sym2 f, ω, s) = ΓR(s)ΓR(s+ 1)ΓR(s+ 2− k − δ)L(Sym2 f, ω, s),
where δ is 0 or 1 according as χ(−1)ω(−1) = 1 or −1 can be continued to a meromorphic
function of s ∈ C, which is holomorphic except for possible simple poles at s = k and
36
s = k − 1. Moreover, the function R(Sym2 f, ω, s) has a pole at s = k if and only if the
following conditions are satised:
1. χω is a non-trivial character of order 2;
2.∫
Φ fg=(τ)kdµ 6= 0, where g(τ) =∑∞
n=1 ω(n)ane2πinτ and Φ is a fundamental domain
for Γ0(NM2) \ H.
Proof. This is a combination of Theorems 1 and 2 of [Shi75].
An identity between Dirichlet series
The following lemma gives an expression for the Euler product of D(f, g, s) in terms
of the roots of f and g.
Lemma 1. Suppose we have formally
∞∑n=1
anns
=∏p
((1− αpp−s)(1− βpp−s))−1,
∞∑n=1
bnns
=∏p
((1− α′pp−s)(1− β′pp−s))−1.
Then
∞∑n=1
anbnns
=∏p
(1−αpβpα′pβ′pp−2s)((1−αpα′pp−s)(1−αpβ′pp−s)(1−βpα′pp−s)(1−βpβ′pp−s))−1.
Proof. See [Shi76][Sec.3, Lemma 1].
If L(s) is a Dirichlet series which has an Euler product expansion, the expression LN (s)
denotes the Dirichlet series with the Euler factors at the primes dividing N removed. For
example,
L(1N , s) = ζN (s),
where L(1N , s) is the Dirichlet series attached to 1N , the trivial Dirichlet character mod N .
37
Proposition 4. Let f ∈ Mk(Γ0(N), χ) be a newform and let ω be a primitive Dirichlet
character of conductor M . Then
LMN (χ2ω2, 2s+ 2− 2k)D(f, fω, s) = LMN (χω, s+ 1− k)L(Sym2 f, ω, s)
for all s ∈ C.
Proof. For <(s) large enough, this follows easily from computing the Euler factors at all
primes p of the four Dirichlet series involved (using Lemma 1, in particular). The result
then follows for all s by Theorem 7.
By the Rankin-Selberg method, the Petersson norm of f can be expressed in terms
of the residue of D(f, fρ, s) at s = k. On the other hand, the last proposition establishes
a relation between D(f, fω, s) and L(Sym2 f, ω, s). In the next two sections, relations
between D(f, fρ, s) and D(f, fω, s), for some ω, are found.
2.3 F Petersson norm of newforms and special values of their symmetricsquare L-functions
This section is based on the argument of [Hid81][Sec.5].
The general case
From now on, suppose that f is a newform. Using the fact that the adjoint of the
Hecke operator Tn under the Petersson inner product is χ(n)Tn, for n coprime to N , one
sees immediately that
an = χ(n)an for gcd(n,N) = 1. (2.7)
It follows that
DN (f, fρ, s) = DN (f, f χ0 , s),
where χ0 is the primitive Dirichlet character attached to χ. Since f is new of level N ,
Equation (2.7) also holds for gcd(n,N ′) = 1, where N ′ is the conductor of χ. It follows
38
that
DN ′(f, fρ, s) = DN ′(f, f
χ0 , s). (2.8)
Let N ′p and Np be the p-part of N ′ and N , respectively (note that N ′p|Np for all primes
p). Then
apap = pk−1 if p|N and Np = N ′p, (2.9)
apap = pk−2 if p|N and Np = p,N ′p = 1, (2.10)
ap = 0 if p2|N and Np = N ′p (2.11)
(see [Hid81][Eqn.5.10]). It follows that the Euler factors of the Dirichlet series in (2.8) dier
only at the primes for which N ′p = Np. This proves the following
Proposition 5. Let f ∈ Sk(Γ0(N), χ) be a newform, let N ′ be the conductor of χ and let
N ′p and Np be the p-parts of N ′ and N , respectively. Then
D(f, fρ, s) =∏
p:N ′p=Np
(1− pk−1−s)−1D(f, f χ0 , s).
Using this, one can now prove the following
Theorem 8. Let f ∈ Sk(Γ0(N), χ) be a newform and let N ′ be the conductor of χ. Then
〈f, f〉 =
(π
2
φ(N)
NN ′φ(N/N ′)
(4π)k
(k − 1)!
)−1
L(Sym2 f, χ0, k),
where φ is the Euler totient function.
Proof. Using Proposition 4 with ψ = χ0, Proposition 5 and by comparing residues at s = k,
one sees that
∏p:N ′p=Np
(1− p−1)ζN (2) ress=kD(f, fρ, s) = ress=1 ζN (s)L(Sym2 f, χ0, k)
39
⇔∏
p|(N/N ′)
(1− p−1)−1∏p|N
(1− p−2)π2
6ress=kD(f, fρ, s) = L(Sym2 f, χ0, k).
Using the Rankin-Selberg method to relate D(f, fρ, s) and 〈f, f〉, we get
∏p|(N/N ′)
(1−p−1)−1∏p|N
(1−p−2))π2
6
3
πN−1
∏p|N
(1 + p−1)−1 (4π)k
(k − 1)!〈f, f〉
= L(Sym2 f, χ0, k)
⇔ π
2
1
N
∏p|N
(1− p−1)∏
p|(N/N ′)
(1− p−1)−1 (4π)k
(k − 1)!〈f, f〉 = L(Sym2 f, χ0, k).
The theorem follows since
φ(N) = N∏p|N
(1− p−1).
The special case where f has trivial character
In the special case where the character of f is trivial, i.e. f ∈ Sk(Γ0(N)), the formula
of Theorem 8 simplies to
〈f, f〉 =
(π
2N
(4π)k
(k − 1)!
)−1
L(Sym2 f, 1, k). (2.12)
In this case, the completed symmetric square L-function
L∗(Sym2 f, 1, s) = (N2)sΓR(s)ΓR(s+ 1)ΓR(s+ 2− k)L(Sym2 f, 1, s)
satises the functional equation
L∗(Sym2 f, 1, 2k − 1− s) = L∗(Sym2 f, 1, s).
This information allows ecient computations with PARI/GP.
40
In particular, this formula applies to the newforms
∆(q) = q∞∏n=1
(1− qn)24,
∆5(q) = q
∞∏n=1
(1− qn)4(1− q5n)4 and
∆11(q) = q∞∏n=1
(1− qn)2(1− q11n)2
in S12(SL2(Z)), S4(Γ0(5)) and S2(Γ0(11)), respectively.
2.4 F Petersson norm of newforms with real Fourier coecients
In the special case where f is twisted by the trivial character, the formula of Proposition
4 gives
LN (χ2, 2s+ 2− 2k)D(f, f, s) = LN (χ, s+ 1− k)L(Sym2 f, 1, s).
Suppose now that f has real Fourier coecients and that χ is not the trivial character
mod N (since this case was treated in the previous section). Then, using the formula of
Theorem 6, one sees that L(Sym2 f, 1, s) has a pole at s = k (since LN (χ, s) is analytic
for χ non-trivial). It then follows from Theorem 7 that χ has order 2. This leads to the
following
Theorem 9. Let f be a newform with real Fourier coecients and non-trivial character.
Then
〈f, f〉 =
(π
2
φ(N)
N2
(4π)k
(k − 1)!
)−1
LN (χ, 1) ress=k L(Sym2 f, 1, s).
Proof. From the previous discussion, we have
ζN (2) ress=kD(f, f, s) = LN (χ, 1) ress=k L(Sym2 f, 1, s).
41
Since f = fρ, Theorem 6 gives
∏p|N
(1− p−2)π2
6
3
πN−1
∏p|N
(1 + p−1)−1 (4π)k
(k − 1)!〈f, f〉
= LN (χ, 1) ress=k L(Sym2 f, 1, s)
⇔ π
2N
∏p|N
(1− p−1)(4π)k
(k − 1)!〈f, f〉 = LN (χ, 1) ress=k L(Sym2 f, 1, s),
from which the result follows.
42
CHAPTER 3Complex multiplication and modular forms over C
The theory of complex multiplication is certainly one of the most beautiful subject in
number theory. It is also vast. For this reason, only the results that are needed in this
thesis are introduced.
Some notation
From now on in this thesis, let K be an imaginary quadratic eld of discriminant
D embedded in C. Moreover, let H be its Hilbert class eld, that is the maximal abelian
unramied extension of K, which has degree hK over K. Let also OK be the ring of integers
of K, IK be the group of fractional ideals of K and ClK be the ideal class group of K.
3.1 CM points in H, L and Y (SL2(Z))
Let a ∈ IK be a fractional ideal of K. Then a is a free Z-module of rank 2 and
a⊗ R = K ⊗ R = C,
so a can be viewed as a lattice in C, i.e. an element of L. This gives a map
(Ω, a) 7→ Ωa : C× × IK −→ L.
The points of L in the image of this map for some K are called CM points.
Note that one can map a fractional ideal a to H in many ways by rst xing an
oriented basis (i.e. by taking a preimage in GL+) and then mapping it to H. Since there is
no canonical choice for this basis, there is no canonical map from IK to H. One may still
43
call a point of H which can be obtained from this procedure a CM point of H. Note that
those are simply the points of H which are algebraic of degree 2 over Q.
3.2 F CM values of modular functions
A central object in the automorphic side of complex multiplication is the modular
function j introduced in the rst chapter. Then one has the following important
Theorem 10. Let a ∈ IK be a fractional ideal of K and let A ∈ ClK be an ideal class of
K. Then
1. j(A) is an algebraic integer in H,
2. [K(j(A)) : K] = [Q(j(A)) : Q] = hK and so K(j(A)) = H,
3. the j(A) are Galois conjugate over K and Q as A ranges over the classes in ClK ,
4. if p is a prime of K, then
j(a)
(H/K
p
)= j(p−1a),
where(H/Kp
)∈ Gal(H/K) is the Artin symbol.
Proof. This is a restatement in terms of ideal classes of [Sil94][Thm.4.3] and [Sil94][Thm.6.1].
See the second part of this thesis for the correspondence between ideal classes and elliptic
curves with complex multiplication.
The values of j at CM points are called singular moduli.
It is a basic fact that modular functions on SL2(Z) can be seen as rational functions
in j with coecients in C, i.e. as element in
C(j).
Then it follows from the above theorem that
44
Corollary 1. Let f be a modular function on SL2(Z) with algebraic Fourier coecients,
i.e. an element of Q(j). Then if f is dened at an ideal class A,
f(A) ∈ Q.
3.3 F Siegel units
Siegel units form a collection of units in the Hilbert class eld of K that will appear
later in the formulas for the Petersson inner product of weight one theta series.
As above, let a be a fractional ideal of K and dene
ϕa =∆(OK)
∆(a−1). (3.1)
This quantity has weight 12 in the ideal a in the sense that
ϕza = z−12ϕa (3.2)
for all z ∈ C×. Then one has the following
Theorem 11. Let a and ϕa be as above. Then
1. ϕa is an algebraic number in H;
2. ϕaOH = a−12OH ;
3. if p is a prime of K, then
ϕ
(H/K
p
)a = ϕapϕ
−1p ,
where(H/Kp
)∈ Gal(H/K) is the Artin symbol.
Proof. See [DS87, Sec.2.2] or [Lan87][Ch.12, Sec.2] (the formulation is slightly dierent
there).
It follows directly from this theorem that
δa = α12ϕhKa = α12
(∆(OK)
∆(a−1)
)hK, (3.3)
45
where αOK = ahK , is a unit in the Hilbert class eld of K which depends only on the ideal
class of a. Those units are called Siegel units.
Let N(a) denote the ideal norm of a. Then N(µa) = |µ|2N(a) for any µ ∈ K and
N(a) =2a(a)√|D|
, (3.4)
where a is the modular form dened in Section 1.3. With α as above, we see that N(a)hK =
|α|2 and so
|δa| = (N(a)6|ϕa|)hK (3.5)
is a unit in H which depends only on the ideal class of a. In fact, one can see using the
second point of the previous theorem that
(N(a)6|ϕa|)2
is also a unit in the Hilbert class eld (see [Lan87][Ch.12, Sec.2]).
3.4 F CM values of modular forms
If f is a modular form of weight greater than 0, the statement "f is algebraic at CM
points" does not make sense. One can salvage this by introducing periods in the statements.
Proposition 6. Let f be a modular form of weight k and level 1 with algebraic Fourier
coecients, and let K be an imaginary quadratic eld. Then there exists a constant Ω(K),
depending only on K, such that
f(a) ∈ Ω(K)kQ
for all fractional ideals a of K (equivalently, for all τ ∈ K ∩H).
Proof. This is [RBvdG+08][Ch.1, Prop.26].
Note that this constant Ω(K) is well-dened only up to an algebraic number. However,
using the Chowla-Selberg formula, one can nd an explicit expression for an Ω(K).
46
Proposition 7. One can choose Ω(K) to be
ΩK =1√
4π|D|
|D|−1∏n=1
Γ
(n
|D|
)χD(n)wK/(4hK)
in the previous proposition.
Proof. This result is proved in [RBvdG+08][Ch.1, Formula (97)], for example, and follows
directly from the Chowla-Selberg formula.
The complex number ΩK is called the Chowla-Selberg period attached to K.
3.5 F CM values of nearly holomorphic modular forms
Recall that the graded C-algebra of nearly holomorphic modular forms for SL2(Z) is
generated by the Eisenstein series E2, E4 and E6. From the previous section, it follows
that the modular forms E4/Ω4K and E6/Ω
6K take algebraic values at fractional ideals of K.
The following proposition proves that a similar statement is true for E2.
Proposition 8. Let E2 be the C∞ Eisenstein series of weight 2 on SL2(Z) and let a be a
fractional ideal of K. Then
E2(a) ∈ Ω2KQ.
Proof. This is [RBvdG+08][Ch.1, Prop.28].
Corollary 2. Let f be a nearly-holomorphic modular form of weight k on SL2(Z) with
algebraic Fourier coecients, i.e. an element of Q[E2, E4, E6], and let a be a fractional
ideal of K. Then
f(a) ∈ ΩkKQ.
47
CHAPTER 4Petersson inner product of theta functions
In this chapter, the Rankin-Selberg method is used to compute the Petersson inner
product of theta series attached to K in terms of L-functions of Hecke characters, C∞
Eisenstein series and derivatives of nearly-holomorphic Eisenstein series.
4.1 F Hecke characters of type A0 with trivial conductor
Let K, OK , etc. be as in the previous chapter. A Hecke character of K of type
A0 with trivial conductor is given by the following data: a pair of integers (k1, k2) and a
homomorphism
ψ : IK → C×
such that
ψ((α)) = αk1αk2
for all α ∈ K×. For simplicity, a Hecke character of type A0 is often represented by the
function ψ. The pair (k1, k2) is called the innity type of ψ. To this pair, one can attach
the function
X : K× → C×
which sends α to αk1αk2 . Note that for ψ to be well dened, X has to be trivial on O×K .
Hecke characters of innity type (0, 0) are called class characters, since they naturally
descend to characters of the class groups ClK = IK/(α) : α ∈ K×. Any two Hecke
characters with the same innity type dier by a class character, since their quotient has
innity type (0, 0). In particular, there are only nitely many Hecke characters of a xed
innity type.
48
Given a class character χ and a function X on K× as above which is trivial on O×K ,
one can obtain a Hecke character by sending a ∈ IK to X(µ)χ(a0), where µ ∈ K×, a0 is an
integral ideal and a = (µ)a0. By the above remarks, all Hecke characters of type A0 can be
obtained in this way. This gives a simple and explicit way to construct Hecke characters,
since the class group can be computed explicitly.
L-functions of Hecke characters
Let ψ be a Hecke character as above and dene
L(ψ, s) =∑a
ψ(a)
N(a)sfor <(s) >
k1 + k2
2+ 1, (4.1)
where the sum is taken over all integral ideals of K. In its region of absolute convergence,
this Dirichlet series has an Euler product expansion
L(ψ, s) =∏p
(1− ψ(p)N(p)−s)−1, (4.2)
where the product is taken over all primes of K.
Theorem 12. Let K be an imaginary quadratic eld of discriminant D and let ψ be a Hecke
character of type A0 with innity type (k1, k2) as above. Dene the completed L-function
attached to ψ as
L∗(ψ, s) = |D|s/2ΓC(s−min(k1, k2))L(ψ, s) for <(s) >k1 + k2
2+ 1,
where ΓC(s) = (2π)−sΓ(s). Then
1. L∗(ψ, s) extends to a meromorphic function on C, with poles at s = 0 and s = 1 if
and only if ψ is the trivial character;
2. L∗(ψ, s) satises the functional equation
L∗(ψ, 1 + k1 + k2 − s) = W (ψ)L∗(ψ, s),
49
where W (ψ) is some complex number of absolute value 1 and ψ is the Hecke character
dened as ψ(a) = ψ(a).
Note that L(ψ, s) is simply the Dedekind zeta function of K when ψ is the trivial
character.
Proof. This follows from [MM06][Thm.3.3.1]. Indeed, it suces to note that
ψ(a) = ξ(a)N(a)k1+k2
2 ,
where ξ is a Hecke character as dened in the reference. Then
L∗(ψ, s) = cΛ
(ξ, s− k1 + k2
2
)for some constant c and the theorem follows.
The number W (ψ) that appears in the functional equation can be computed explicitly
in terms of Gauss sums (see [MM06][Sec. 3.3]).
4.2 F Theta functions attached to imaginary quadratic elds
From now on, the term Hecke character will always refer to Hecke characters of type
A0 and conductor 1, as they were dened in the previous section.
Theta functions attached to Hecke characters of conductor one
Let ψ be such a character of innity type (k1, k2) and consider the following q-expansion
θψ(q) = cψ +∑a
′ψ(a)qN(a), (4.3)
where the sum is taken over all non-trivial integral ideals of K and
cψ =
0 if ψ is non-trivial
hKwK
if ψ is the trivial character
.
50
For θψ to be the q-expansion of a modular form, it is necessary that k1 = 0 or k2 = 0. To
see this, suppose that θψ was a modular form. Then its L-series would be
L(θψ, s) = L(ψ, s) (4.4)
and then one sees that the gamma factor of L(ψ, s), namely Γ(s − min(k1, k2)), can be
the gamma factor of a modular form only if min(k1, k2) = 0. Furthermore, by looking
at the exponential factor and the weight of L(ψ, s), one sees that θψ would have weight
max(k1, k2) + 1 and level |D|. Finally, by computing ap(θψ) and by looking at the Euler
factor at p of L(θψ, s), one can deduce that θψ would have character the Kronecker symbol
χD dened as
χD(p) =
1 if p splits in K
−1 if p is inert in K
0 if p is inert in K
.
As it turns out, the condition k1 = 0 or k2 = 0 is also sucient for θψ to be a modular
form.
Note that there is no loss of generality in supposing that ψ has innity type (k1, 0),
since then ψ has innity type (0, k1) and the equality θψ = θρψ implies that
θψ = θψ,
since θρψ = θψ. To prove the last equality, compute ap(θψ) and note that
ψ(p) = ψ(p)
for any prime ideal p in K since
ψ(p)ψ(p) = N(p)k1 = ψ(N(p)OK) = ψ(pp) = ψ(p)ψ(p).
51
From now on, let ψ is a Hecke character of innity type (2`, 0) for some ` ≥ 0 (note that
k1 + k2 must be even when the conductor is OK , since ±1 ∈ O×K).
Theorem 13. Let ψ be a Hecke character of innity type (2`, 0) for some ` ≥ 0. Then
θψ(q), given by (4.3), is the q-expansion of a modular form
θψ ∈M2`+1(Γ0(|D|), χD),
where χD is the Kronecker symbol introduced above. If ` 6= 0, the modular form θψ is a
newform. If ` = 0, the modular form θψ is a newform unless ψ2 is the trivial class character,
in which case it is an Eisenstein series.
The class characters of order dividing 2 are called genus characters and the Eisenstein
series θψ attached to them are called genus Eisenstein series.
To prove Theorem 13, it is convenient to introduce another collection of theta series.
Theta functions attached to ideals
Let a be a fractional ideal of K, let ` ≥ 0 be an integer as above and consider the
following q-expansion
θa,`(q) =∑λ∈a
λ2`qN(λ)/N(a). (4.5)
Note that there is no loss of generality in supposing that a is integral, since
θµa,`(q) = µ2`θa,`. (4.6)
Note also that
θρa,` = θa,`. (4.7)
The following result is well-known.
52
Theorem 14. Let a be an integral ideal of K and let ` be a positive integer. Then θa,`(q),
given by (4.5), is the q-expansion of a modular form
θa,` ∈M2`+1(Γ0(|D|), χD),
where χD is the Kronecker symbol. Moreover, the modular form θa,` is a cusp form if and
only if ` > 0.
Proof. First, note that a is a free Z-module of rank 2 and soN(λ)/N(a) is a binary quadratic
form which has integral coecients. Second, a direct computation shows that λ2` is a
spherical polynomial for this quadratic form. The result then follows from the general
theory of classical theta series (see [Iwa97, Thm.10.9], for example).
The relation between the theta functions attached to ideals and those attached to
Hecke characters is given in the following
Proposition 9. Let ψ be a Hecke character of innity type (2`, 0) and let a1, . . . ahK be a
set of integral ideal class representatives of ClK . Then
θψ(q) =1
wK
hK∑j=1
ψ−1(aj)θaj ,`(q)
and
θa,` =wKhK
∑ψ
ψ(a)θψ,
where the sum is over the hK Hecke characters of innity type (2`, 0).
Proof. Let a be an integral ideal. Then since the a−1j are also representatives of ClK , one
can write a = λa−1j for some unique j and some λ ∈ aj which is unique up to a unit.
Therefore
θψ(q) =1
wK
hK∑j=1
∑λ∈aj
ψ(λa−1j )qN(λ)/N(aj) =
1
wK
hK∑j=1
ψ−1(aj)∑λ∈aj
λ2`qN(λ)/N(aj).
53
The second equality follows from this equality and the following orthogonality relation:
∑ψ
ψ(a) =
λ2`hK if a = λOK for some λ ∈ K
0 otherwise. (4.8)
To prove this, x a Hecke character ψ0 of innity type (2`, 0) and note that
∑ψ
ψ/ψ0(a) =
hK if a = λOK for some λ ∈ K
0 otherwise,
by the orthogonality relation for characters of the abelian group ClK .
It is now possible to prove Theorem 13.
Proof. (of Theorem 13) It is clear from the modularity of the θa,` and Proposition 9 that
the θψ are modular forms of the correct weight, level and character.
For ` 6= 0, the cuspidality of the θa,` also implies that of θψ. Moreover, the fact that
L(θψ, s) has an Euler product of the right shape proves that θψ is a newform (see [DS05,
Thm.5.9.2], which is a partial converse to Theorem 5).
For ` = 0, a ner analysis is needed to prove that θψ is an Eisenstein series if and only
if ψ2 is trivial. See [Kan12, Thm.14], for example. Once this is known, the same argument
as above proves that θψ is a newform when it is cuspidal.
The identities in the previous proposition can also be used to obtain the well-known
decomposition of L-functions of Hecke characters in terms of Eisenstein series evaluated at
CM points. To see this, rst note that
L(θa,`, s) = N(a)sG2`,s−2`(a). (4.9)
Then
L(θψ, s) =1
wK
hK∑j=1
ψ−1(aj)L(θaj ,`, s)
54
and so
L(ψ, s) =1
wK
hK∑j=1
ψ−1(aj)N(aj)sG2`,s−2`(aj). (4.10)
The space of theta functions
By Proposition 9, the modular forms θa,` and θψ span the same space insideM2`+1(Γ0(|D|), χD)
as a ranges over all fractional ideals of K and ψ ranges over the Hecke characters of innity
type (2`, 0). Let ΘK,` ⊆M2`+1(Γ0(|D|), χD) denote this space.
When ` = 0, relation (4.7) becomes
θa,0 = θa,0. (4.11)
Letting Z/2Z act on the set θa1,0, . . . , θahK ,0 via ρ, where the aj are representatives of
ClK , one sees using Burnside's Lemma that this set contains (hK + gK)/2 orbits, where
gK = [ClK : Cl2K ] is the number of genera.
This relation (4.11) and relation (4.6) are the only relations between theta functions,
for any ideal and any ` ≥ 0. More formally:
Proposition 10. Let ` ≥ 0 and let a1, . . . , ahK be representatives of ClK . Then
if ` = 0, the C-vector space ΘK,` has dimension (hK + gK)/2;
if ` 6= 0, the C-vector space ΘK,` has dimension hK .
Proof. This is well-known and follows from the classical theory of integral binary quadratic
forms. See [Kan12, Prop.7], for example.
One can then prove the following
Corollary 3. Let ψ1, . . . , ψgK be the gK genus characters of ClK . Then the Eisenstein
series θψj for j = 1, . . . , gK are C-linearly independent. Moreover, the space spanned by
the hK − gK cusp forms θψ, for ψ a class character of order greater than 2, has dimension
(hK − gK)/2.
55
4.3 F Petersson inner product of theta functions attached to imaginary qua-dratic elds
In this section, the Rankin-Selberg method is used to nd formulas for the Petersson
norm of cuspidal theta functions.
Symmetric square L-function of theta functions attached to Hecke charac-ters
Recall that the Petersson norm of a newform is related to a special value or to the
residue of a twist of its symmetric square L-function. For theta functions, this L-function
is related to the L-function of a Hecke character.
Proposition 11. Let ψ be a Hecke character of innity type (2`, 0) for some ` ≥ 0 and let
ω be a primitive Dirichlet character. Then
L(Sym2 θψ, ω, s) = LD(ω, s− 2`)∑a
ω(N(a))ψ2(a)
N(a)s,
where the sum is taken over all integral ideals of K.
Proof. For <(s) large enough, all Dirichlet series involved have an Euler product expansion
and the statement follows from a straightforward calculation.
Corollary 4. The following equalities hold:
L(Sym2 θψ, 1, s) = ζD(s− 2`)L(ψ2, s)
and
L(Sym2 θψ, χD, s) = L(χD, s− 2`)LD(ψ2, s).
Proof. The rst equation is clear and the second follows from the fact that χD(N(p)) is 1
if p divides a prime of Q which is unramied in K and 0 otherwise.
56
Petersson norm of cuspidal theta functions attached to Hecke characters
Let ψ be a Hecke character of innity type (2`, 0) which is not a genus character. Then
θψ is a newform and θρψ = θψ, so one can apply either the formula of Theorem 9:
〈θψ, θψ〉 =
(π
2
φ(|D|)D2
(4π)2`+1
Γ(2`+ 1)
)−1
L(χD, 1) ress=2`+1 L(Sym2 θψ, 1, s) (4.12)
or the general formula of Theorem 8:
〈θψ, θψ〉 =
(π
2
φ(|D|)D2
(4π)2`+1
Γ(2`+ 1)
)−1
L(Sym2 θψ, χD, 2`+ 1).
For no particular reason, let us use the rst formula. Of course, the second gives the same
result.
Proposition 12. Let ψ be a Hecke character of innity type (2`, 0) which is not a genus
character. Then
〈θψ, θψ〉 =4hKwK
√|D|Γ(2`+ 1)
(4π)2`+1L(ψ2, 2`+ 1).
Proof. Using the previous Corollary, one sees that
ress=2`+1 L(Sym2 θψ, 1, s) = ress=1 ζD(s)L(ψ2, 2`+ 1) =∏p|D
(1− p−1)L(ψ2, 2`+ 1).
The formula then follows from a simple calcuation, using (4.12) and the class number
formula
L(χD, 1) =2πhK
wK√|D|
.
Note that if ` = 0 and ψ is a genus character, the above formula for 〈θψ, θψ〉 is not
well-dened since in this case L(ψ2, s) is the Dedekind zeta function of K which has a pole
at s = 1.
57
Using (4.10) and the previous proposition, it follows that
L(ψ2, s+ 2`+ 1) =1
wK
hK∑j=1
ψ−2(aj)N(aj)s+2`+1G4`,s−2`+1(aj).
If ` 6= 0, the Eisenstein series G4`,s−2`+1 is well-dened at s = 0 and so
L(ψ2, 2`+ 1) =1
wK
hK∑j=1
ψ−2(aj)N(aj)2`+1G4`,−2`+1(aj).
Using Equation (1.76) and (3.4), one can compute
δ2`−12 E2(a) = 2−1(2πi)−2δ2`−1
2 G∗2,0(a)
= 2−1(2πi)−4`G∗4`,−2`+1(a)
= (2π)−4` 1
4(2`)!
(√|D|N(a)
2π
)−2`+1
G4`,−2`+1(a)
= (2π)−1−2` 1
4(2`)!(
√|D|N(a))−2`+1G4`,−2`+1(a),
and so
L(ψ2, 2`+ 1) =4(2π)2`+1
√|D|2`−1
wK(2`)!
hK∑j=1
ψ−2(aj)N(aj)4`δ2`−1
2 E2(aj). (4.13)
If ` = 0, Theorem 1 says that the Eisenstein series G∗0,s+1, and hence G0,s+1, has a
pole at s = 0. Using Kronecker's rst limit formula, equation (3.4) and the orthogonality
relation for the non-trivial character ψ2, one sees that
L(ψ2, 1) = − π
3wK√|D|
hK∑j=1
ψ−2(aj) log(N(aj)6|∆(aj)|). (4.14)
All those computations essentially prove the following
58
Theorem 15. Let ψ be a Hecke character of innity type (2`, 0) which is not a genus
character. Then if ` 6= 0,
〈θψ, θψ〉 = (|D|/4)`4hKw2K
∑A∈ClK
ψ2(A)δ2`−12 E2(A),
while if ` = 0,
〈θψ, θψ〉 = − hK3w2
K
∑A∈ClK
ψ2(A) log(N(A)6|∆(A)|).
Proof. When ` > 0, rst note that
N(a)4`δ2`−12 E2(a) = δ2`−1
2 E2(a−1)
and that
ψ(a)δ2`−12 E2(a)
quantity depends only on the ideal class of a. The formula for 〈θψ, θψ〉 follows then from
Proposition 12 and (4.13).
When ` = 0, it suces to use Proposition 12 and (4.14), since
N(a−1)6|∆(a−1)| = N(a)6|∆(a)|
and since this quantity depends only on the ideal class of a.
Note that it follows from those formulas that
〈θψ, θψ〉 = 〈θψ−1 , θψ−1〉
when ` = 0, as expected (since θψ = θψ−1 in this case).
Corollary 5. Let ψ be a Hecke character of innity type (2`, 0) which is not a genus
character and let ΩK be the Chowla-Selberg period attached to K as in Proposition 7. Then
〈θψ, θψ〉 ∈ Ω4`KQ.
59
Proof. This follows from the previous theorem and the corollary to Proposition 8 applied
to the nearly holomorphic modular form δ2`−12 E2.
Obtaining the case ` = 0 from the case ` > 0
The formula of Theorem 15 for ` > 0 does not make sense for ` = 0 since it involves
the expression
δ−12 E2.
However, if this expression made sense, it would have to be a C∞ modular form of weight
zero such that
δ0F = E2.
We claim that
δ0 log(=(τ)6|∆(τ)|) = −12E2(τ),
where
δ0 =1
2πi
d
dτ.
This follows from the well known fact (see [RBvdG+08, Prop. 7]) that
d
dqlog ∆(q) =
1
2πi
d
dτlog ∆(τ) = 1− 24
∞∑n=1
σ1(n)qn. (4.15)
Indeed, since
log |∆(τ)| = <(log ∆(τ)),
it implies thatd
dτlog |∆(τ)| = 1
2
d
dτlog ∆(τ)
(since dFdτ = dF
dτ = 0 if F (τ) is holomorphic). Letting
δ−12 E2(a) = − 1
12log(N(a)6|∆(a)|) + C,
60
where C is some constant, in the formula of Theorem 15 for ` > 0 gives exactly the formula
for ` = 0.
This relation between E2 and the logarithmic derivative of ∆ is the starting point of
the calculations that will be done in the second part of this thesis.
Petersson inner product of theta functions attached to ideals
When ` > 0, one can use Proposition 9 and Theorem 15 to prove the following
Proposition 13. Let ` > 0 and let a and b be two fractional ideals of K. Then
〈θa,`, θb,`〉 = 4(|D|/4)`∑
abc2=λcOK
λ2`c δ
2`−12 E2(c),
where the sum is over a set of representatives c of ideals classes in ClK such that abc2 =
λcOK .
Then one has the following
Corollary 6. Let ` > 0 and let a and b be two fractional ideals of K which are not in
the same genus, i.e. their classes in ClK dier mod Cl2K . Then θa and θb are orthogonal
under the Petersson inner product.
Proof. The sum in the above expression for 〈θa,`, θb,`〉 is empty.
This last corollary can be proved without the explicit formulas of the previous propo-
sition. However, it is not clear how one could prove the following corollary without those
formulas.
Corollary 7. Let ` > 0 and let a, b and c be fractional ideals of K. Then
〈θca,`, θcb,`〉 = N(c)2`〈θa,`, θb,`〉.
Proof. Clear from the previous proposition.
61
CHAPTER 5Petersson norm of weight one theta functions
In weight one, one can use the formula of Theorem 15 to obtain more algebraic infor-
mation about the Petersson norm of theta series attached to class characters.
5.1 F Petersson norm of theta functions and Siegel units
Recall (see (3.3) and (3.5)) that the absolute value of Siegel units is given by
|δa| = (N(a)6|∆(OK)/∆(a−1)|)hK .
When ψ is a class character which is not a genus character, one can rearrange the formula
in Theorem 15 to get
〈θψ, θψ〉 =1
3w2K
hK∑j=1
ψ2(aj) log |δaj |. (5.1)
Note that wK is always equal to 2 with these assumptions on ψ.
5.2 F On Stark's observation
Throughout this section, let K = Q(√−23) and as before let H denote the Hilbert
class eld of K. Recall Stark's observation in the introduction on the relation between the
Petersson norm of a theta series and the special value at s = 1 of an Artin L-function. He
observed that the Petersson norm of the weight one theta function attached to a degree 2
Artin representation of Gal(H/Q) ' S3 has Petersson norm
3 log ε,
where ε is a real root of x3 − x− 1, which generates the Hilbert class eld of K (over K).
62
With our notation, the theta series considered by Stark is θψ, where ψ is one of the
two non-trivial class characters of K. Its Petersson norm could be expressed as a linear
combination of logarithms of Siegel units as in (5.1), but to obtain Stark's result one has
to work directly from the formula of Theorem 15. Then
〈θψ, θψ〉 = −1
4
3∑j=1
ψ2(aj) log(N(aj)6|∆(aj)|) = −3
3∑j=1
ψ2(aj) log(√N(aj)|η(aj)|2),
where
η(q) = exp(2πiτ/24)∏n=1
(1− qn)
is the Dedekind eta function. As is well known, the function |η(τ)|2 on H is a C∞ modular
form of weight (0, 1) (this follows from [DS05, Prop.1.2.5], for example), so it has a well-
dened value on lattices in L. Dene
Φ(a) =√N(a)|η(a)|2.
Then Φ depends only on the ideal class of a and has the property that
Φ(a) = Φ(a) = Φ(a−1).
Moreover, (Φ(OK)
Φ(a−1)
)12hK
= |δa|. (5.2)
With this notation,
〈θψ, θψ〉 = −33∑j=1
ψ2(aj) log Φ(a−1j ) = 3 log
3∏j=1
Φ(a−1j )−ψ
2(aj).
Letting a be any non-principal ideal of K, we see that
〈θψ, θψ〉 = 3 logΦ(a−1)
Φ(OK).
63
This computation uses the fact that
ψ2(a) + ψ2(a−1) = 2<(ζ3) = −1,
where ζ3 is any non-trivial third root of unity. As one can verify numerically,
Φ(a−1)
Φ(OK)
is a real root of x3 − x− 1, so we recover Stark's example.
5.3 F On generalisations of Stark's observation
Based on the above computation and the formula
〈θψ, θψ〉 = hK log
hK∏j=1
Φ(aj)−ψ2(aj), (5.3)
it might seem reasonable to conjecture that
κψ =
hK∏j=1
Φ(aj)−ψ2(aj) (5.4)
is a unit in the Hilbert class eld of K. Note that the individual numbers Φ(a) are usually
transcendental and that ψ2(a) is a root of unity. In fact, when ψ is a genus character κψ
can be computed explicitly in terms of the Chowla-Selberg period dened in Proposition
7 (actually, this statement is equivalent to the Chowla-Selberg formula, see [RBvdG+08,
Sec.6.3]). It follows from those observations that the algebraicity of κψ depends crucially
on the character ψ.
Generalisation to class number 3
There are exactly 16 imaginary quadratic elds of class number 3, the one with the
largest discriminant in absolute value being Q(√−907) (see [Wat04, Table 4]). In this case,
64
the same computation as above shows that
〈θψ, θψ〉 = 3 logΦ(a−1)
Φ(OK),
where ψ is one of the two non-trivial class characters of K and the following lemma shows
that κψ is a unit.
Lemma 2. Let K be an imaginary quadratic eld and let a be an ideal of K. Then
Φ(a−1)
Φ(OK)
is an algebraic integer and a unit.
Proof. It suces to note that (Φ(a−1)
Φ(OK)
)12hK
= |δa|.
When K has class number 3, it appears numerically that the unit Φ(a−1)/Φ(OK) is in
fact in the Hilbert class eld of K. This gives numerical evidence that Stark's observation
generalizes to all imaginary quadratic elds of class number 3.
Generalisation to class number 5
When K is one of the 25 imaginary quadratic elds of class number 5 (see [Wat04,
Table 4]), Stark's observation does not seem to generalize. In other words, the complex
number κψ does not seem to be a unit for any class character ψ.
The main dierence between the class number 3 and class number 5 case is that the
later involves non-trivial 5th roots of unity, which do not have rational real part.
The general case
The above observations lead to the following
65
Proposition 14. Let ψ be a character of the class group of K as above and suppose that ψ2
is a non-trivial character with rational real part. Then κψ is an algebraic integer which is a
unit. If ψ2 is a non-trivial genus character corresponding to the factorization D = D1D2,
with D1 > 0, then
κψ = ε
4hD1hD2
wKwD2D1
,
where εD1 is a fundamental unit of Q(√D1), hDj is the class number of Q(
√Dj) and wD2
is the number of roots of unity in Q(√D2).
Proof. Since ψ2 is non-trivial,
κψ =
hK∏j=1
Φ(aj)−ψ2(aj) =
hK∏j=1
(Φ(aj)
Φ(OK)
)−ψ2(aj)
.
The rst claim follows from the previous lemma and the observation that(Φ(a)
Φ(OK)
)−ψ2(a)( Φ(a)
Φ(OK)
)−ψ2(a)
=
(Φ(a)
Φ(OK)
)−2<(ψ2(a))
if a is not equivalent to a in ClK and(Φ(a)
Φ(OK)
)−ψ2(a)
=Φ(a)
Φ(OK)
otherwise.
For the second part, rst note that
L(ψ2, s) = L(χD1 , s)L(χD2 , s),
where χDj is the Kronecker symbol attached to the discriminantDj (see [Coh07, Prop.10.5.19]).
Then
L(ψ2, 1) = L(χD1 , 1)L(χD2 , 1) =4πhD1hD2
wD2
√|D|
log εD1
66
by the class number formula and so by Proposition 12
hK log κψ = 〈θψ, θψ〉 =4hK
√|D|
wK(4π)L(ψ2, 1) = hK
4hD1hD2
wKwD2
log εD1 .
Note that the power of εD1 which appears in the above proposition is not always
integral. For example, with D = −39, D1 = 13 and D2 = −3, the proposition gives
κψ = ε1313. This example also proves that κψ is not always in the Hilbert class eld (since
H/K has degree 4 in this case).
It also follows from that proposition that κψ, does not always generate the Hilbert
class eld of K (even when it is in H).
Since ψ2 has rational real part if and only if it has order dividing 4 of 6, it follows that
κψ is a unit for some ψ whenever gcd(hk, 6) > 1. We believe that the converse is true.
Conjecture 1. If K has class number coprime to 6, then κψ is not algebraic for any ψ.
Numerical evidence for this conjecture is given in Chapter 12.
67
Part II
p-adic Interpolation
68
CHAPTER 6p-adic modular forms
In the next few sections, test objects, weight space and Tate curves will be introduced
over rings, leading to the denition of algebraic modular forms and generalized p-adic
modular forms. Our treatment of the theory of algebraic and generalized p-adic modular
forms follows closely Chapters II and V of [Kat76], respectively. The rst section is there
to motivate the passage from the classical analytic theory (formulated in terms of points
in the upper half-plane and lattices) to the algebraic theory (to be formulated in terms of
moduli spaces of enhanced elliptic curves). The fact that the algebraic theory of modular
forms is equivalent over C to the theory presented in Chapter 1 will lead to an algebraic
interpretation of the complex formulas of the rst part. One can then use the powerful
theory of p-adic modular forms to interpolate them.
6.1 Towards an algebraic theory of modular forms
Throughout this section, the notation is the same as in Chapter 1.
The idea of the algebraic theory of modular forms is to interpret modular forms as
functions on moduli spaces of elliptic curves with extra structure. To see why this idea is
not far fetched, let L ∈ L be a lattice. Then it is well-known that the quotient space C/L
is a smooth Riemann surface of genus one, i.e. a torus. One can also explicitly realize this
space as the set of complex points of an elliptic curve over C via the map
φL : C/L −→ EL(C)
z 7−→ [℘L(z) : ℘′L(z) : 1],
69
where ℘L is the Weierstrass ℘ function attached to the lattice L and
EL : Y 2 = 4X3 − g2(L)X − g3(L),
with
g2 = 60G4 and g3 = 140G6.
(see [Sil09, Coro.5.1.1]). This illustrates the fact that, loosely speaking, the set of lattices
L can be thought of as a set of elliptic curves over C.
To view modular forms as functions of elliptic curves, one should rst turn this last
statement into an actual bijection by nding a way to attach a lattice to an elliptic curve
E over C. The way one can do this depends on how this elliptic curve E is presented. If E
is dened by a Weierstrass equation
EA,B : Y 2 = 4X3 +AX +B,
then it is well known (see [Sil09, Thm.5.1]) that there exists a unique lattice L(A,B) such
that
g2(L(A,B)) = −A and g3(L(A,B)) = −B.
However, if E is given geometrically as a nonsingular curve over C with a distinguished
point, there is no canonical way to attach a Weierstrass equation to it. To solve this
problem, recall that one can recover L from the space C/L by integrating the translation
invariant dierential dz along the elements of the singular homology group H1(C/L,Z):
L =
∫γdz : γ ∈ H1(C/L,Z)
(see [Sil09, Prop.5.6]). Letting
L(E,ω) =
∫γω : γ ∈ H1(E(C),Z)
70
for any elliptic curve E over C and any invariant dierential ω on E, it follows that
L(EL, ωL) = L,
where
ωL = dXY .
This proves that there is a sequence of bijections of sets
L ∼→Y 2 = 4X3 +AX +B : A3 − 27B2 6= 0
∼→ (E,ω) / ≈,
where (E,ω) ≈ (E′, ω′) if there is a isomorphism φ : E∼→ E′ such that φ∗(ω′) = ω. A
couple (E,ω) is called a framed elliptic curve.
Now let f be a modular form of weight k and level one. By denition, it satises
f(λL) = λ−kf(L)
for all λ ∈ C× and L ∈ L. It is a simple exercise to see that if L ∈ L corresponds to the
framed elliptic curve (E,ω), the lattice λL corresponds to the framed elliptic curve (E, λω).
Viewing f as a function on framed elliptic curves, this is saying that
f(E, λω) = λ−kf(E,ω).
At the level of Weierstrass equations, the change from L to λL corresponds to the change
of variables
X = λ2X ′ and Y = λ3Y ′.
This proves that the weight of a modular form can be recovered from the corresponding
function on framed elliptic curves.
71
One can also recover the q-expansion from this function on framed elliptic curves. To
see this, rst note that for xed τ0 ∈ H, one has the following sequence of equalities in C
f(τ0) = f([τ0, 1])
= f (C/[τ0, 1], dz)
= f
(Y 2 = 4X3 − g2(τ0)X − g3(τ0),
dX
Y
)= f
(Y 2 = 4X3 − g2(q0)X − g3(q0),
dX
Y
),
where q0 = e2πiτ0 ∈ H. The Weierstrass equation
Y 2 = 4X3 − g2(q)X − g3(q)
denes an elliptic curve over C((q)) (use (1.22)). Applying the change of variables
X = (2πi)2X ′ and Y = (2πi)3Y ′
gives an isomorphism(Y 2 = 4X3 − g2(q)X − g3(q),
dX
Y
)≈(Y ′2 = 4X ′3 − (2πi)−4g2(q)X ′ − (2πi)−6g3(q),
1
2πi
dX ′
Y ′
)and the last elliptic curve is dened over Q((q)). In fact, one can do even better by letting
X ′ = x+1
12and Y ′ = x+ 2y,
which gives an isomorphism(Y ′2 = 4X ′3 − (2πi)−4g2(q)X ′ − (2πi)−6g3(q),
dX ′
Y ′
)≈(y2 + xy = x3 +B(q)x+ C(q),
dx
x+ 2y
),
72
where
B(q) = −5∞∑n=1
n3qn
1− qn
C(q) = − 1
12
∞∑n=1
(7n5 + 5n3)qn
1− qn.
Note that this last elliptic curve is now dened over Z((q)) (in fact, it has coecients in
Z[[q]]). Its discriminant is
∆(q) = q∞∏n=1
(1− qn)24.
It follows from this last sequence of change of variables that
f(q0) = (2πi)kf
(y2 + xy = x3 +B(q0)x+ C(q0),
dx
x+ 2y
).
One might hope that
f(q) = (2πi)kf
(y2 + xy = x3 +B(q)x+ C(q),
dx
x+ 2y
), (6.1)
where now q is seen as a variable. However, this equality does not make sense yet, since f
is only dened on lattices and the elliptic curve
y2 + xy = x3 +B(q)x+ C(q)
is not dened over C (so one cannot attach a lattice to it). This idea still leads to a good
denition of the q-expansion and one can indeed make sense of this equality.
73
6.2 Test objects and trivialized elliptic curves
Let R be a ring (commutative with 1). An elliptic curve over R is dened as a smooth
and proper morphism
E
Spec(R)
whose geometric bres are connected curves of genus one, together with a section
E
Spec(R)
e
HH .
Let E be an elliptic curve over a ring R and let N ≥ 1 be a positive integer. The kernel
of multiplication by N is a nite at commutative group scheme over R of rank N2, which
is denoted E[N ]. It is étale if and only if N is invertible in R.
6.2.1 Level N test objects
As above, let N ≥ 1 be a positive integer and let E/R be an elliptic curve over R. As
in [Kat76, Sec.2.0], a Γ(N)arith level structure is dened as an isomorphism
β : µN × Z/NZ −→ E[N ]
of group schemes over R such that the Weil pairing on E[N ] corresponds to the canonical
pairing
〈(ζ1, n), (ζ2,m)〉 =ζm1ζn2
on µN × Z/NZ. A Γ(N)arith-test object over R is a triple
(E/R, ω, β),
where
74
E/R is an elliptic curve over R;
ω is a nowhere vanishing invariant dierential on E;
β is a Γ(N)arith level structure
(see [Kat76, Sec.2.1]).
6.2.2 Trivialized elliptic curves
A p-adic ring B is dened as a ring which is complete and separated with respect to
its p-adic topology (e.g. Zp or Fp, but not Qp).
Let E/B be an elliptic curve over B. As in [Kat76, Sec.5.0], a trivialization of E is
dened as an isomorphism
ϕ : E −→ Gm
of formal groups over B. Note that a trivialization ϕ of E/B is equivalent to a compatible
sequence of inclusions
µpr → E
for r ≥ 0. In particular, trivialized elliptic curves are bre-by-bre ordinary (note that the
residue eld of every maximal ideal of B is of characteristic p). A Γ(N)arith level structure
β on E/B is said to be compatible with the trivialization ϕ if the composition of maps
µprβ→ E
ϕ−→ Gm
is the inclusion, where pr||N and Gm is the multiplicative group scheme. A trivialized
Γ(N)arith elliptic curve over B is a triple
(E/B, ϕ, β),
where
E/B is an elliptic curve over B;
ϕ is a trivialization of E;
75
β is a compatible Γ(N)arith level structure
(see [Kat76, Sec.5.1]).
6.3 Weight and Tate curves
Throughout this section, let N ≥ 1 be a positive integer, and let (E/R, ω, β) and
(E/B, ϕ, β) be as above.
6.3.1 Weight
Let f be a map from the set of Γ(N)arith-test objects over R to R. For λ ∈ R×, dene
the map [λ]f as
[λ]f(E,ω, β) = f(E, λ−1ω, β). (6.2)
Then f is said to be of weight k ∈ Z if
[λ]f = λkf for all λ ∈ R×. (6.3)
Note that in level 1 over C one recovers the notion of weight introduced in Section 1.2.
Now let f be a map from the set of trivialized Γ(N)arith elliptic curves over B to B.
In this case it is not as straightforward to dene the weight of f , since there is no obvious
action of B× on trivialized Γ(N)arith elliptic curves. However, one can let Z×p act on a
trivialization ϕ : E −→ Gm by viewing a ∈ Z×p as a compatible sequence of elements in
(Z/prZ)× for r ≥ 1. If (E/B, ϕ, β) is a trivialized Γ(N)arith elliptic curves over B, then so
is
(E/B, a−1ϕ, β (a, a−1)).
More generally, following [Kat76, Sec.5.3], let
G(N) =
(a, b) ∈ Z×p × (Z/NZ)× : a ≡ b (mod pr), where pr||N
(6.4)
76
and for (a, b) ∈ G(N), dene the map [a, b]f as
[a, b]f(E/B, ϕ, β) = f(E/B, a−1ϕ, β (b, b−1)). (6.5)
Then f is said to be of weight κ, where κ is a continuous character
κ : G(N) −→ B×,
if
[a, b]f = κ(a, b)f for all (a, b) ∈ G(N). (6.6)
If κ = κk · χ, where
κk(a, b) = ak
for some k ∈ Z and χ is a nite order character of G(N), then f is said to be of weight k
and character (or nebentypus) χ.
6.3.2 Tate curves
In Section 6.1, it was shown that the elliptic curve
Tate(q) : y2 + xy = x3 +B(q)x+ C(q) (6.7)
is dened over Z((q)). This elliptic curve is called the Tate curve of level one. Note also
that Tate(q) is equipped with a canonical dierential
ωcan =dx
x+ 2y,
so that
(Tate(q), ωcan) (6.8)
77
is a level one test object (see [Kat76, Sec.2.2]). Using the same notation as in Section 6.1,
one has the following isomorphisms of test objects:
(C/[τ0, 1], dz) ≈ (C/2πi[τ0, 1], 2πidz) ≈(C×/qZ0 ,
dt
t
),
where the rst map is z 7→ 2πiz, the second is z 7→ e2πiz, and t = e2πiz is the parameter on
C×. Therefore, one can think of Tate(q) as "Gm/qZ". In level N , the relevant Tate curve,
denoted Tate(qN ), is given by the Weierstrass equation
Tate(qN ) : y2 + xy = x3 +B(qN )x+ C(qN ) (6.9)
and can be thought of as "Gm/qNZ". With that in mind, it is clear that the N -torsion of
Tate(qN ) should correspond to the N2 points
ζiNqj for 0 ≤ i, j ≤ N − 1,
where ζN is a primitive Nth root of unity. This gives a canonical Γ(N)arith level structure
βcan : µN × Z/NZ −→ Tate(qN )
and so
(Tate(q), ωcan, βcan) (6.10)
is a Γ(N)arith-test object (see [Kat76, Sec.2.2]). Over the ring Zp((q)), the p-adic completion
of Zp((q)), one can also trivialize the Tate curve Tate(qN ) in a canonical way (the formal
completion of Gm/qNZ is Gm). Since this trivialization, denoted ϕcan, is compatible with
βcan, one obtains trivialized Γ(N)arith elliptic curves
(Tate(qN ), ϕcan, βcan)
(see [Kat76, Sec.5.2.0]).
78
6.4 Algebraic and generalized p-adic modular forms
As in the previous section, let N ≥ 1 be an integer.
6.4.1 Algebraic modular forms
Let R be a ring and let A be an R-algebra. Following [Kat76, Sec.2.1], an algebraic
modular form dened over R of weight k ∈ Z and level Γ(N)arith is a rule f which assigns
to every Γ(N)arith-test object
(E/A, ω, β)
over A an element
f(E/A, ω, β) ∈ A
and which satises the following properties:
The value f(E/A, ω, β) depends only on the isomorphism class of (E/A, ω, β);
If R′ is a R-algebra, the formation of f commutes with base change from R to R′.
Symbolically,
f((E/A, ω, β)⊗R R′) = f(E/A, ω, β)⊗R 1
inside A ⊗R R′, where (E/A, ω, β) ⊗R R′ is the test object obtained from (E/A, ω, β)
by base change from R to R′;
For every λ ∈ A×,
f(E/A, λ−1ω, β) = λkf(E/A, ω, β).
The R-module of algebraic modular forms is denoted
V algk (R,Γ(N)arith).
As in [Kat76, Sec.2.2], the q-expansion of an algebraic modular form of level Γ(N)arith
is dened as
f(Tate(qN ), ωcan, βcan) ∈ R⊗ Z((q)) ⊆ R((q)).
79
The very important q-expansion principle says that the map
V algk (R,Γ(N)arith) −→ R((q)) (6.11)
which sends algebraic modular forms to their q-expansion is injective (see [Kat76, Sec.2.2.6]).
Moreover, if R ⊆ R′, then
V algk (R,Γ(N)arith) → V alg
k (R′,Γ(N)arith)
and an element f ∈ V algk (R′,Γ(N)arith) lies in V alg
k (R,Γ(N)arith) if and only if its q-
expansion lies in R((q)) (see [Kat76, Sec.2.2.6]).
Finally, note that there is an action of diamond operators on V algk (R,Γ(N)arith) as
follows: for b ∈ (Z/NZ)× dene
〈b〉f(E/A, ω, β) = f(E/A, ω, β (b, b−1)) (6.12)
for f ∈ V algk (R,Γ(N)arith) (see [Kat76, Sec.5.4.8]). If
〈b〉f = χ(b)f
for some homomorphism χ : (Z/NZ)× −→ R×, then f is said to be of character (or
nebentypus) χ.
6.4.2 Generalized p-adic modular forms
Let B be a p-adic ring, let A be a p-adic ring which is also a B-algebra. Following
[Kat76, Sec.5.1], a generalized p-adic modular form dened over B of level Γ(N)arith is
dened as a rule f which assigns to every trivialized Γ(N)arith elliptic curve
(E/A, ϕ, β)
80
over A an element
f(E/A, ϕ, β) ∈ A
and which satises the following properties:
The value f(E/A, ω, β) depends only on the isomorphism class of (E/A, ω, β);
If B′ is p-adic ring which is also a B-algebra, the formation of f commutes with base
change from B to B′.
The B-module of such objects is denoted
V gen(B,Γ(N)arith).
Let
κ : Z×p −→ B×
be a continuous homomorphism. Then, as in [Kat76, Sec.5.3], a generalized p-adic modular
form is said to be of weight κ if
[a, b]f(E/A, ω, β) = κ(a, b)f(E/A, ω, β)
for every (a, b) ∈ G(N). The B-module of generalized p-adic modular forms of weight κ is
denoted
V genκ (B,Γ(N)arith).
As in [Kat76, Sec.5.1], the q-expansion of a generalized p-adic modular form is dened
as
f(Tate(qN ), ϕcan, βcan) ∈ B((q)).
The q-expansion principle says that the map
V genκ (B,Γ(N)arith) −→ B((q)) (6.13)
81
which sends generalized p-adic modular forms to their q-expansion is injective (see [Kat76,
Sec.5.2.1]). Moreover, if B ⊆ B′, then
V genκ (B,Γ(N)arith) → V gen
κ (B′,Γ(N)arith)
and an element f ∈ V genκ (B′,Γ(N)arith) lies in V gen
κ (B,Γ(N)arith) if and only if its q-
expansion lies in B((q)) (see [Kat76, Sec.5.2.2]).
6.5 Comparison between modular forms
6.5.1 Classical and algebraic modular forms over C
The correspondence between elliptic curves over C and lattices in C, presented in
Section 6.1, gives a map
V algk (C,Γ(1)arith) −→M !
k(Γ(1)),
where M !k(Γ(1)) denotes the space of weakly holomorphic modular forms of level 1 (i.e. the
space of modular forms which are only required to have meromorphic q-expansion at the
cusps (see Section 1.3)). A priori, there is no map in the other direction, since one cannot
attach a lattice in C to every elliptic curve over a C-algebra. As it turns out, this map is
bijective and generalizes to higher level (see [Kat76, Sec.2.4]). However, this natural map
does not quite preserve q-expansions1 . For this reason, consider instead the normalized
map which sends
f ∈ V algk (C,Γ(N)arith)
1 In [Kat76, Sec.2.4], Katz mentions that this map is a q-expansion preserving bijection.This is because his denition of the q-expansion of a classical modular form diers from ours.We prefer the denition given here since it is the most widely used. For example, accordingto Katz's denition, the Eisenstein series G4(τ) =
∑m,n(mτ + n)−4 has q-expansion with
rational coecients.
82
to the function
fan : GL+ −→ C
dened as
fan(ω1;ω2) = (2πi)−kf
(C/[ω1, ω2], dz, β(e2πia/N , b) =
aω1 + bω2
Nmod L
)(6.14)
(note the power of 2πi). Then fan is Γ(N) invariant and homogeneous of weight k. More-
over, one has the following
Proposition 15. Let R be a subring of C. Then the above map f 7→ fan induces a q-
expansion preserving bijection between V algk (R,Γ(N)arith) and the set of weakly holomorphic
modular forms in M !k(Γ(N)) whose q-expansion coecients are in R.
In particular, if the q-expansion of a classical modular form f is dened over R and
the elliptic curve C/[ω1;ω2] is dened over R (i.e. g2(ω1;ω2), g3(ω1;ω2) ∈ R), then
(2πi)kf(ω1;ω2) ∈ R.
Proof. The rst part of the proposition is a restatement of [Kat76, Sec.2.4]. To prove the
second statement, note that if f ∈ Mk(Γ(N)) corresponds to falg ∈ V algk (R,Γ(N)arith),
then
(2πi)kf(ω1;ω2) = falg (C/[ω1;ω2], dz, β) ∈ R,
by the denition of algebraic modular forms.
By dening
M !k(R,Γ(N)) = f ∈M !
k(Γ(N)) : f(q) ∈ R((q)), (6.15)
one can interpret this proposition as saying that the map
V algk (R,Γ(N)arith) −→M !
k(R,Γ(N))
83
dened above is a q-expansion preserving bijection.
6.5.2 Algebraic and generalized p-adic modular forms
The following proposition says that there is a q-expansion preserving map
V algk (B,Γ(N)arith) −→ V gen(B,Γ(N)arith)
which also preserves the nebentypus (see [Kat76, Sec.5.4] for a denition of the map).
Proposition 16. Let B be a p-adic ring and let f ∈ V algk (B,Γ(N)arith) be an algebraic
modular form of nebentypus χ : (Z/NZ)× −→ B×. Then the corresponding generalized
p-adic modular form has weight k and nebentypus χ, i.e. it has weight κ = κk · χ0, where
χ0 : G(N) −→ B× sends (a, b) to χ(b) and κk(a, b) = ak.
Proof. See [Kat76, Lem.5.4.10].
6.6 p-adic Eisenstein series
Recall (see (1.24)) that the Eisenstein series Ek has q-expansion
Ek(q) = −Bk2k
+∞∑n=1
σk−1(n)qn for k ≥ 4,
and so it follows directly from Proposition 15 that there exist algebraic modular forms with
those q-expansions. In weight 2, the Eisenstein series
E2(τ) =1
8π=(τ)− 1
24+
∞∑n=1
σ1(n)qn
of Section 1.4 is not holomorphic on H and so one cannot apply the same reasoning.
However, one has the following
Proposition 17. Let k ≥ 2. Then there exists a p-adic modular form
Ek ∈ V genk (Zp,Γ(1))⊗Qp
84
such that
Ek(Tate(q), ϕcan, βcan) = −Bk2k
+∞∑n=1
σk−1(n)qn.
Proof. For k > 2, this follows from the above discussion and Proposition 16.
For k = 2, the proof is more involved and the reader is referred to [Kat76, Lem.5.7.8]
for an algerbo-geometric denition of E2.
Note in particular that the p-adic modular form E2 ∈ V gen2 (Zp,Γ(1)) ⊗ Qp has q-
expansion
E2(q) = − 1
24+∞∑n=1
σ1(n)qn
This abuse of notation (using the symbol Ek to denote classical and p-adic Eisenstein
series) should not cause any confusion; the Eisenstein series considered will be clear from
the context and the expression E2(q) will always be equal to the last equation.
6.7 Changing the level at p
It follows from the previous results that most of the modular forms considered so
far can be viewed as generalized p-adic modular forms. From now on, generalized p-adic
modular forms will be simply called p-adic modular forms.
The following proposition is characteristic of the p-adic theory of modular forms.
Proposition 18. Let B be a p-adic ring, let N0 ≥ 1 be an integer coprime to p and let
r ≥ 0. Then there is a G(prN0) ∼= G(N0) equivariant, q-expansion preserving isomorphism
V gen(B,Γ(prN0)arith) ∼= V gen(B,Γ(N0)arith).
Proof. See the discussion below and [Kat76, Sec.5.6] for a proof.
85
To see why such an isomorphism should exist, consider the following example. First,
it follows from the results of Section 1.5 that the classical modular form E(p)k dened as
E(p)k (q) = Ek(q)− pk−1VpEk(q) = (1− pk−1)
−Bk2k
+
∞∑n=1
σ(p)k−1(n)qn,
where σ(p)k−1(n) =
∑p-d|n d
k−1, has weight k for Γ0(p), hence also for Γ(p). For most primes
p, there exists a p-adic modular form in
V genk (Zp,Γ(p)arith)
with q-expansion E(p)k (q). By the previous proposition, it follows that E(p)
k can be seen
as a p-adic modular form of weight k and level Γ(1)arith. In this case, one can prove this
directly. To do so, let ki be a sequence of even integers which tend to ∞ in R and to k in
G(1) = Z×p . Then for any n ≥ 1, one sees that
limi→∞
σki−1(n) = σ(p)k−1(n)
in Qp. Moreover, it is also true that
limi→∞
(1− pki−1)−Bkiki
= limi→∞
ζp(1− ki) = ζp(1− k) = (1− pk−1)ζ(1− k),
by the continuity of the Kubota-Leopoldt p-adic L-function. It follows that
limi→∞
Eki(q) = E(p)k (q)
in Zp[[q]]⊗Qp. Viewing each Eki(q) as an element of
V gen(Zp,Γ(1)arith)⊗Qp
and using the q-expansion principle, it follows that E(p)k can also be viewed as a p-adic
modular form of level 1.
86
6.8 Operators on generalized p-adic modular forms
In the classical case, the operator Vp sends a modular form
f(q) ∈Mk(Γ0(N))
to the modular form
Vpf(q) = f(qp) ∈Mk(Γ0(pN)).
In the p-adic theory of modular forms,
V genκ (B,Γ(N)arith) ∼= V gen
κ (B,Γ(pN)arith),
and so it seems like a p-adic analogue of the Vp operator, if it exists, should preserve the
level p-adically. This is indeed the case.
Proposition 19. Let N ≥ 1 be an integer and let B be a p-adic ring. There exists an
endomorphism Vp of
V genκ (B,Γ(N)arith)
which acts on q-expansions as Vpf(Tate(qN ), ωcan, βcan) = f(Tate(qpN ), ωcan, βcan) (i.e.
Vpf(q) = f(qp)).
Proof. See [Kat76, Sec.5.5].
The operator Vp is also called the Frobenius operator, since in characteristic p (e.g. if
B = Fp) it satises the equation
Vpf(q) = f(qp) = f(q)p.
87
6.9 Theta operator on generalized p-adic modular forms
In the classical case, recall that the Shimura-Maass operator was dened in (1.71) as
δkf(τ) =1
2πi
(∂f(τ)
∂z+kf(τ)
τ − τ
)on weight k modular forms. Recall also that it increases the weight by 2 and does not
preserve holomorphicity. However, it preserves the ring of nearly holomorphic modular
forms, which in level one is simply
C[E2, E4, E6].
The analogue of this dierential operator in the p-adic theory is Serre's theta operator,
which will be denoted DN .
Proposition 20. Let N ≥ 1 be an integer. There exists an endomorphism DN of
V gen(Zp,Γ(N)arith)
which acts on q-expansions as q ddq , i.e. it makes the following diagram commutative:
V gen(Zp,Γ(N)arith)DN //
V gen(Zp,Γ(N)arith)
Zp((q))q ddq
// Zp((q))
,
where Zp((q)) is the p-adic completion of Zp((q)). Moreover, it is of weight 2, in the sense
that
[a, b] DN = a2DN [a, b]
for all (a, b) ∈ G(N).
88
Proof. The proof involves an analysis of the Gauss-Manin connection on the DeRham co-
homology of the universal elliptic curve over V gen(Zp,Γ(N)arith) and will not be given here.
See [Kat76, Sec.5.8].
89
CHAPTER 7Complex multiplication from an algebraic point of view
Recall that in this text, K denotes an imaginary quadratic eld and OK denotes its
ring of integers.
In Chapter 3, the analytic theory of complex multiplication was introduced. In parti-
cular, CM points in L were dened as the set of lattices of the form
Ωa,
where Ω ∈ C× and a is a fractional ideal in K. One may wonder if the associated elliptic
curve
C/Ωa
has special properties. To answer this, rst note that if
C/L ∼−→ E(C) and C/L′ ∼−→ E′(C),
then any isogeny φ : E −→ E′ over C corresponds to a unique λ ∈ C× such that
λL ⊆ L′
(see [Sil09, Thm.4.1, Ch.VI]). In particular,
EndC(C/L) = λ ∈ C : λL ⊆ L
and so
EndC(C/Ωa) ⊇ OK .
90
In other words, the elliptic curve C/Ωa has extra endomorphisms (other than the trivial
ones).
Conversely, one can show that if L ∈ L and λ ∈ C× are such that
λL ⊆ L,
then λ is an element in an order in a quadratic imaginary eld. Therefore the endomorphism
ring of elliptic curves over C is either Z or an order in an imaginary quadratic eld. In
the later case, the elliptic curve is called a CM elliptic curve or is said to have complex
multiplication.
It follows from the above discussion that the set of elliptic curves over C with CM by
the maximal order OK of an imaginary quadratic eld K is in bijection with the CM points
of L corresponding to K (as dened in Chapter 3). It also follows that there are exactly
hK isomorphism classes (over C) of elliptic curves with CM by OK . See [Sil94, Ch.II] for a
detailed exposition of the theory of CM elliptic curves.
7.1 F CM values of algebraic modular forms
The rst step in passing from the classical theory of complex multiplication to the
algebraic one was to go from lattices in C to framed elliptic curves. The next step is to
analyse the eld of denition of CM elliptic curves.
Proposition 21. Let E be an elliptic curve over C with CM by the maximal order OK .
Then
1. E is isomorphic over C to an elliptic curve dened over H (the Hilbert class eld of
K);
2. Suppose that E is dened over a number eld F . Then every endomorphism of E is
dened over FK, the compositum of F and K.
91
Proof. The rst point follows from the fact that the j-invariant of CM elliptic curves is an
algebraic integer in the Hilbert class eld of K (see Proposition 10).
Then second point is proved in [Sil94, Ch.II, Thm.2.2], for example.
Now let (E/F , ω, β) be a Γ(N)arith-test object over a number eld F and suppose
that E has complex multiplication by OK . Suppose also that the complex multiplication
endomorphisms are dened over F . Then one has the following
Proposition 22. Let f ∈Mk(F,Γ(N)) be a classical modular form with Fourier coecients
in F and let (E/F , ω, β) be a Γ(N)arith-test object as above. Then
(2πi)2a+2b+kδa2b+kEb2f(ω1;ω2) ∈ F
for any a, b ≥ 0, where (ω1;ω2) is the point of LN = Γ(N) \GL+ corresponding to the test
object (E/F , ω, β) base changed to C.
Note that this Proposition could equivalently have been stated in terms of algebraic
modular forms.
Proof. If a = b = 0, this follows directly from the correspondence between algebraic and
classical modular forms and the denition of algebraic modular forms.
The main diculty, and the only place where the hypothesis of complex multiplication
is used, is in proving that
(2πi)2E2(ω1;ω2)
lies in F . This is proved in [Kat76, Thm.4.0.4] (note that Katz's function S corresponds to
the function 24(2πi)2E2 in our notation).
In particular, one has the following
Corollary 8. Let f ∈ F [E2, E4, E6] be a nearly holomorphic modular form of weight k with
Fourier coecients in a number eld F containing the Hilbert class eld H of K, and let
92
a be a fractional ideal of K. Then there exists a complex number Ωa depending only on a
such that
f(a) ∈ (2πiΩa)−kF.
Proof. The elliptic curve
C/a
is isomorphic over C to an elliptic curve (E/F , ω) dened over F (since F contains the
Hilbert class eld of K). Let L ∈ L be the lattice corresponding to the complexication of
E/F . Then there exists Ωa ∈ C× such that
a = ΩaL.
The claim follows from the equality
(2πiΩa)kf(a) = (2πi)kf(L)
and the previous proposition.
This Corollary may look stronger than Proposition 6, since the number eld in which
f takes values at CM points is known. However, this is not really the case, since the extra
ambiguity in Proposition 6 (coming from the fact that the values were only required to be
algebraic) allowed us to choose the period ΩK explicitly (using the closed formula in the
Chowla-Selberg formula) and independently of the fractional ideal a.
7.2 CM values of p-adic modular forms
As above, let (E/F , ω, β) be Γ(N)arith-test object with CM. In this section, x a prime
P of F and let p be the rational prime which it divides in OF . In order to talk about
"CM values" of p-adic modular forms, one must make a few assumptions about the test
objects. First, suppose that the test object (E/F , ω, β) has good reduction at P, in the
93
sense that there exists a Γ(N)arith-test object over the ring OF,P of P-integers of OF which
gives (E/F , ω, β) after extension of scalars from OF,P to F . Second, suppose that E has
ordinary reduction at P (note that this hypothesis is equivalent to p being split in K).
Given such a test object (E/F , ω, β) and an embedding of F in C, one can attach an ele-
ment (ω1;ω2) ∈ LN to its complexication. On the other hand, one can consider it as a test
object (E/FP, ω, β) over the P-adic completion FP of F at P. On the modular forms side,
any algebraic modular form f ∈Mk(F,Γ(N)arith) can be considered as a classical modular
form with Fourier coecients in F or as p-adic modular form in V genk (Zp,Γ(N)arith)⊗ FP.
Then one has the following
Theorem 16. With the hypotheses and notations as above, the quantity
DaNE
b2f(E/FP
, ω, β) ∈ FP
actually lies in F and is equal to the quantity
(2πi)2a+2b+kδa2b+kEb2f(ω1;ω2) ∈ F.
Proof. If a = b = 0, this is clear.
The main diculty is to compare the complex value
(2πi)2E2(ω1;ω2)
with the P-adic value
E2(E/FP, ω, β),
where now E2 is the p-adic modular form dened in Section 6.6. For a proof of this, see
[Kat76, Thm.8.0.9].
94
CHAPTER 8p-adic interpolation of Petersson inner product of theta series
In this chapter, a few p-adic measures are constructed, including one with values in the
p-adic ring of generalized p-adic modular forms and one with values in a subring of Cp. The
second measure naturally gives rise to a Cp-valued analytic function on the p-adic weight
space
W = Homcont(Z×p ,Z×p ).
By evaluating this function at a point outside the range of interpolation, namely at κ = −1,
a p-adic analogue of the Petersson inner product of two weight one theta series attached
to ideals is found. This result can be interpreted as a p-adic analogue of Kronecker's rst
limit formula.
8.1 Review of p-adic integration theory
Part of the material of this section is contained in [Kat76, Sec.6.0]. Other, more
detailed, references are [Col04, Sec.1.4] and [MSD74, Ch.7].
As before, let B be a p-adic ring and let
G = lim←−n
G/Gn,
where
G1 ⊇ · · · ⊇ Gn · · · ⊇ ∩Gn = 1,
be a pronite group (e.g. Zp or Z×p ).
95
Dene a B-valued measure on G as a B-linear map
µ : C0(G,B) −→ B,
where C0(G,B) is the ring of B-valued continuous functions on G. Note that µ is auto-
matically continuous (C0(G,B) is equipped with the p-adic topology). The value of µ on
f ∈ C0(G,B) is denoted ∫Gfdµ or
∫Gf(g)dµ(g).
Equivalently, one could dene a B-valued measure as a nitely additive map µ on com-
pact open subsets of G, i.e. as a B-valued distribution on G. Indeed, any such distribution
extends uniquely to a B-linear map
µ : LC(G,B) −→ B,
where LC(G,B) is the ring of locally constant B-valued functions on G. Using the density
of LC(G,B) in C0(G,B), one can then show that µ extends uniquely to a B-valued measure
on G.1
8.1.1 Integration over Zp
When G = Zp, one has a lot more information about the set of B-valued measure on
Zp, thanks to the structure theorem for C0(Zp,Zp).
1 Note that B is simply assumed to be a p-adic ring and so it is not necessarily proniteor equipped with an absolute value. In particular, one cannot dene p-adic measures asbounded p-adic distributions, as in the classical theory. However, the density of LC(G,B)in C0(G,B) and the extension property can still be proved using only the fact that B iscomplete and separated with respect to its p-adic topology.
96
Theorem 17 (Mahler's Theorem). Any f(x) ∈ C0(Zp,Zp) can be uniquely written as
f(x) =∞∑n=0
an(f)
(x
n
),
where (x
n
)=x(x− 1) . . . (x− n+ 1)
n!for all n ≥ 0
and
an(f) =n∑i=0
(−1)i(n
i
)f(n− i) −→
n−→∞0 in Zp.
Conversely, for any sequence an of p-adic integers tending to 0 p-adically, the series
f(x) =∞∑n=0
an
(x
n
)converges for x ∈ Zp and denes an element f(x) ∈ C0(Zp,Zp) which is such that
an(f) = an for all n ≥ 0,
where an(f) is dened as above.
Proof. This theorem is contained in many places. See [Col04, Thm.1.3.2], for example.
By tensoring with B over Zp and using the fact that
C0(Zp, B) = C0(Zp,Zp)⊗ZpB = C0(Zp,Zp)⊗Zp B,
one has the following corollary (this is [Kat76, 6.0.3]).
Corollary 9. Suppose that B is at over Zp (equivalently, B is of characteristic 0). Then
any f(x) ∈ C0(Zp, B) can be uniquely written as
f(x) =
∞∑n=0
an(f)
(x
n
),
97
where
an(f) =n∑i=0
(−1)i(n
i
)f(n− i) −→
n−→∞0 in B,
and conversely (as above).
This structure theorem for C0(Zp, B) allows one to formally attach the power series
Aµ(T ) =
∫Zp
(1 + T )xdµ(x) =∞∑n=0
(∫Zp
(x
n
)dµ(x)
)Tn ∈ B[[T ]]
to any B-valued measure µ on Zp. This map is sometimes called the Amice transform.
In fact, the Amice transform is more than a formal link between measures and power
series. To illustrate that, let z ∈ 1 + pB and consider the series
fz(x) =∞∑n=0
(z − 1)n(x
n
).
By Mahler's theorem, this series denes a B-valued continuous function on Zp, which is
such that
fz(k) = zk for all k ∈ Z≥0.
Because fz satises the above interpolation property, fz(x) is denoted zx. Using this nota-
tion, one has the following
Lemma 3. Let B be a p-adic ring which is at over Zp, let z ∈ 1+pB and let µ ∈ D0(Zp, B).
Then
Aµ(z − 1) =
∫Zpzxdµ(x).
Proof. This is a generalization of [Col04, Lem.1.4.3]. By the normal convergence of the
series dening the function zx (i.e. for any M ∈ Z≥0, there exists an index nM such that∑∞n=nM
(z−1)n(xn
)∈ pMB), one can interchange the sum and the integral in the expression∫
Zp
∞∑n=0
(z − 1)n(x
n
)dµ(x).
98
Now suppose that B is at over Zp and for every integer n ≥ 0 dene ck(n) ∈ Zp for
1 ≤ k ≤ n in such a way that (x
n
)=
n∑k=0
ck(n)xk.
It is then clear that a measure µ on Zp is determined by its moments
mk(µ) =
∫Zpxkdµ, for k ∈ Z≥0.
Conversely, one has the following
Lemma 4. Let B be a p-adic ring which is at over Zp and let mkk≥0 be a sequence of
elements of B. Then there exists a measure µ ∈ D0(Zp, B) with moments m0,m1, . . . if
and only ifn∑k=0
ck(n)mk ∈ B
for every n ≥ 0. If it exists, the measure is unique.
Proof. This is [Kat76, Lem.6.0.9].
8.1.2 Integration over Z×p
When p is an odd prime, the natural exact sequence
1 −→ Γ −→ Z×p −→ F×p = µp−1 −→ 1,
where µp−1 is the group of p−1 roots of unity and Γ = 1+pZp, gives a canonical isomorphism
Z×p ' µp−1 × Γ. (8.1)
If x ∈ Z×p , let ω(x) and 〈x〉 denote the images of x under the projection to the rst and
second factors of the isomorphism (8.1), respectively.
99
Recall (Subsection 6.3.1) that the weight of a generalized p-adic modular form of level
one with coecients in B is by denition a continuous character
κ : Z×p −→ B×.
The space of all such weights
W(B) = Homcont(Z×p , B×)
is called the weight space (with values in B). In the case where B = Zp, this space is simply
denoted W.
Using the decomposition (8.1), one sees that any element of W is of the form
κi,s(x) = ωi(x)〈x〉s
for some i ∈ Z/(p− 1)Z and s ∈ Zp. It follows that W decomposes as a direct product of
p− 1 copies of Zp, indexed by the characters of µp−1:
κi,s 7→ (i, s) :W ∼−→ (Z/(p− 1)Z)× Zp. (8.2)
In fact, this decomposition also respects the group operations on both sides. The natural
injection from Z into W, which sends the integer k to the character
κk(x) = κk,k(x) = xk,
100
then corresponds to the natural map from Z into the product decomposition (8.2). This
injection induces a topology on Z which makes it dense in W.2 The weights in the image
of Z are called classical weights.
Similarly to the case of Zp, the space W(B) decomposes as a product of copies of B×
indexed by
µ∗p−1(B) = Hom(µp−1, B×).
Now let
µ ∈ D0(Z×p , B)
be a measure on Z×p . By mapping W to C0(Z×p , B), one obtains a continuous function on
weight space
Lµ :W −→ B
whose values at integers k ≥ 0 are the moments of µ:
Lµ(k) = Lµ(κk) =
∫Z×pxkdµ.
Then one has the following
Proposition 23. Let µ ∈ D0(Z×p , B) be a p-adic measure, let i ∈ Z/(p − 1)Z and let
u = 1 + p. Then there exists a unique measure Γ(i)µ ∈ D0(Zp, B) on Zp such that∫
Z×pωi(x)〈x〉sdµ(x) =
∫ZpusydΓ(i)
µ (y)
for all s ∈ Zp.
2 This induced topology is the one where two integers are close if they are p-adically closeand congruent mod p− 1.
101
Proof. This is a straightforward generalization of the proof of [Col04][Prop.1.5.9] to the
case of measures with values in p-adic rings.
The measure Γ(i)µ is sometimes called Leopoldt's gamma transform. In the notation of
the previous proposition, one sees using Lemma 3 that∫ZpusydΓ(i)
µ (y) = AΓ(i)µ
(us − 1), for all s ∈ Zp.
It follows that the function Fµ has a power series expansion in s on each of the p − 1
components of W. Indeed, for each xed i ∈ Z/(p− 1)Z one has
Lµ(κi,s) =
∫Z×pωi(x)〈x〉sdµ(x) = A
Γ(i)µ
(us − 1),
which can be expressed as a power series in s (since us − 1 = sv + s2v2/2 + . . . , where
v = log u). Such functions on Zp (or W) are called p-adic analytic.
To summarize, p-adic measures on Z×p naturally give rise to continuous functions on
weight space, which in turn can be viewed as a nite collection of p-adic analytic functions
on Zp which interpolate the moments of the measure.
8.2 Construction of measures with values in the ring of generalized p-adicmodular forms
From now on, suppose that p is dierent from 2 and 3 and let V denote the p-adic ring
of generalized p-adic modular forms of level one:
V = V gen(Zp,Γ(1)arith).
8.2.1 The measure µE on Zp with values in V
One has the following
102
Lemma 5. There exists a unique measure µE on Zp with values in V whose moments are
given by ∫ZpxkdµE = Dk
1E2 for all k ≥ 0.
Proof. By Lemma 4, it suces to show that
n∑k=0
ck(n)Dk1E2 ∈ V
for all n ≥ 0. But an easy computation using the q-expansion of Dk1E2 shows those p-adic
modular forms have q-expansions
∞∑m=1
(m
n
)σ(m)qm if n > 0
and
E2(q) = − 1
24+∞∑m=1
σ1(m)qm if n = 0,
which are in V (i.e. their q-expansions have coecients in Zp). Note that the assumption
p 6= 2, 3 is needed when n = 0.
The calculation done in the proof of this lemma can be used to prove the following
Proposition 24. Let µE be the measure of Lemma 5 and let f ∈ C0(Zp, V ). Then∫ZpfdµE = −f(0)
24+∞∑n=1
f(n)σ(n)qn.
Proof. The proof of Lemma 5 shows that the statement is true for the functions(xn
)∈
C0(Zp, V ). The result then follows from Corollary 9, the fact that
a0(f) = f(0)
and the continuity of µE .
103
8.2.2 Restriction of µE to Z×p
Since integration over Z×p gives rise to analytic functions on weight space, it is desirable
to restrict µE to Z×p . Dene
µ[p]E ∈ D
0(Z×p , V )
as ∫Z×pg(x)dµ
[p]E (x) =
∫Zpg0(x)dµE(x) =
∫Zpg0(x)χZ×p (x)dµE(x),
where g ∈ C0(Z×p , V ) and g0 ∈ C0(Zp, V ) is g extended by 0 on pZp. Then one has the
following
Proposition 25. Let µ[p]E be the measure dened above. Then∫
Z×pxkdµ
[p]E = Dk
1E[p]2 ,
where the p-adic modular form E[p]2 with q-expansion
E[p]2 (q) =
∑(n,p)=1
σ(n)qn
is the p-depletion of E2(q).
Proof. This follows directly from the denition of µ[p]E and Proposition 24.
That E[p]2 is a p-adic modular form of level Γ(1)arith follows formally from the previous
proposition, since µ[p]E is a p-adic measure with values in V . One can also prove directly the
following
Lemma 6. As in the previous proposition, let
E[p]2 (q) =
∑(n,p)=1
σ(n)qn.
Then if
104
E2(τ) =1
8π=(τ)− 1
24+∞∑n=1
σ(n)qn
is the nearly holomorphic modular form of Chapter 1, one has the equality
E[p]2 (q) = E2(τ)− (p+ 1)VpE2(τ) + pV 2
p E2(τ),
where q = e2πiτ . Moreover, E[p]2 (q) is a holomorphic modular form in M2(Γ0(p2)).
E2(q) = − 1
24+
∞∑n=1
σ(n)qn
is the p-adic modular form of Chapter 6, one has the equality
E[p]2 (q) = E2(q)− (p+ 1)VpE2(q) + pV 2
p E2(q).
Moreover, E[p]2 (q) is a p-adic modular form in V gen
2 (Zp,Γ(1)arith).
Proof. For the rst point, note that the non-holomorphic parts in the right hand side of
the equation cancel. The claim then follows from a computation on q-expansions and the
general fact that the Vp operator sends weight k functions on SL2(Z) to weight k functions
on Γ0(p2).
The second point follows from the same computation on q-expansions and the fact that
the Vp operator preserves the level p-adically.
More generally, one sees that ∫Z×pxκdµ
[p]E = Dκ
1E[p]2 ,
where Dκ1 is dened as
Dκ1E
[p]2 (q) =
∑(n,p)=1
nκσ(n)qn
105
and xκ = κ(x) denotes the image of x ∈ Z×p under κ ∈ W. It then follows from the theory
exposed above that the measure µ[p]E on Z×p induces a p-adic analytic function
Lµ[p]E
:W −→ V,
which is dened as
Lµ[p]E
(κ) = Dκ1E
[p]2 . (8.3)
8.3 Construction of a measure with values in Cp
Our goal in this section is to evaluate the p-adic modular forms Lµ[p]E
(κ) at CM points
and to relate those values to the CM values of the Shimura-Maass derivatives of E2 which,
as was shown in the rst part of this thesis, intervene in the formulas for the Petersson
inner product of theta series. In order to do so, one has to attach trivialized Γ(1)arith
elliptic curves to ideals in quadratic elds.
8.3.1 Trivialized elliptic curves attached to CM elliptic curves
This is well explained and done in more generality in Section 8.3 of [Kat76], so we only
recall the main idea here.
As usual, let K be an imaginary quadratic eld and let H be its Hilbert class eld.
From now on, let a be a fractional ideal of K and suppose the p splits as
pOK = pp
in K. 3
3 Asking for p to split in K is a big restriction, since only half of the prime in Q do so.However, this restriction is necessary since the CM elliptic curves attached to K are notordinary at the inert primes of K, which means that they cannot be trivialized. Since thedierential operator D1 does not preserve overconvergence in general, it is impossible toevaluate all the p-adic modular forms Dk
1E[p]2 (for k ∈ Z≥0) at non-ordinary elliptic curves.
106
Fix an isomorphism
ϕ : Qp/Zp∼−→⋃n≥1
p−na/a (8.4)
and let
H(p∞) =⋃n≥1
H(pn),
where H(pn) is the ray class eld mod pn over K.
By CM theory, the couple (a, ϕ) determines a complex embedding of H(p∞). 4 Fix
a place p∞ of H(p∞) which divides p and let Op∞ be its valuation ring. Then the elliptic
curve over C determined by a has a model Ea over H ∩ Op∞ . Suppose that the action of
OK on H0(Ω1) is compatible with the inclusion OK → H → C.
Over H ∩ Op∞ , there is a canonical splitting
⋃n≥1
ker(pn) =⋃n≥1
ker(pn)×⋃n≥1
ker(pn).
Taking the Cartier dual of ϕ and using the Weil pairing on pn torsion, one obtains an
isomorphism
ϕ :⋃n≥1
ker(pn) =⋃n≥1
p−na/a∼−→ µp∞ (8.5)
Out of the above data, one can extract at least two things. First, the isomorphism
ϕ−1 × ϕ : µp∞ ×Qp/Zp∼−→⋃n≥1
ker(pn) (8.6)
4 The part of CM theory used to prove such a statement was not covered in this thesis.The missing result is that, roughly speaking, the pn torsion points of elliptic curves withCM by OK generate the extension H(pn).
107
gives rise to Γ(pn)arith-level structures on Ea for every n ≥ 0. Second, passing over to Op∞ ,
the isomorphism ϕ becomes equivalent to a trivialization
ϕ : Ea∼−→ Gm.
In this way, to any couple (a, ϕ) one can attached trivialized Γ(pn)arith elliptic curves
(Ea, ϕ, βn)/Op∞
for all n ≥ 0.
8.3.2 The measure on Z×p with values in Cp
Let (a, ϕ) be as above. By evaluating generalized p-adic modular forms at the triviali-
zed elliptic curve attached to (a, ϕ), one obtains a measure valued in the p-adic ring Op∞ .
For reasons that will become clear later, it is preferable to view our measures as valued in
p-adic modular forms of level p2 rather than 1. This leads to the denition of
Lµ[p]E
((a, ϕ), κ) :W −→ Op∞
as the composition
W −→ C0(Z×p ,Z×p )µ[p]E−→ V gen(Zp,Γ(1)arith)
∼−→ V gen(Zp,Γ(p2)arith)eval−→ Op∞ .
This function can be seen as a Cp-valued p-adic analytic function on W.
Our next goal is to relate the values of this function at classical weights to the CM
values of some classical modular form, using Theorem 16. In order to do so, it is necessary
to go from trivialized elliptic curves to elliptic curves equipped with dierentials. As in the
previous subsection, we follow [Kat76, Sec.8.3]. Let
(Ea, ϕ, βn)/Op∞
108
be a trivialized elliptic curve over Op∞ . Then the trivialization
ϕ : Ea −→ Gm
gives rise to the dierential ϕ∗(dT/1 + T ) on Ea dened over Op∞ (see [Kat76], right
before paragraph 8.3.16, for more details). Since Ea is dened over H ∩ Op∞ , there exists
Ωp(a) ∈ O×p∞ such that
Ωp(a)ϕ∗(dT/1 + T )
is dened over Op∞ . Using the complex embedding of Op∞ , one can also nd ΩC(a) ∈ C×
such that
Ωp(a)ϕ∗(dT/1 + T ) = ΩC(a)dz.
The period lattice of Ea is then ΩC(a)a.
Theorem 18. Let k ∈ W be a classical weight. Then using the notation introduced above,
one has the following equality in H
Lµ[p]E
((a, ϕ), k)
Ωp(a)2k+2=
δkE[p]2 (a)
((2πi)−1ΩC(a))2k+2. (8.7)
Proof. After all this work, the proof is a relatively simple computation using the previous
calculations and Theorem 16:
Lµ[p]E
((a, ϕ), k) = Dk1E
[p]2 (Ea, ϕ, β2)
/Op∞
= Ωp(a)2k+2Dk1E
[p]2 (Ea,Ωp(a)ϕ∗(dT/1 + T ), β2)/H∩Op∞
= Ωp(a)2k+2(2πi)2k+2δk2E[p]2 (ΩC(a)a)
= Ωp(a)2k+2(2πi)2k+2ΩC(a)−(2k+2)δk2E[p]2 (a).
Note that the fact that E[p]2 is a holomorphic modular form of weight 2 and level Γ0(p2) is
used when Theorem 16 is applied.
109
8.4 p-adic interpolation of Petersson inner product of theta series
Our goal in this section is to see how one could p-adically interpolate the quantities
δk2E2(a),
where a is some fractional ideal of K, which are directly related via Proposition 13 to the
Petersson inner product of theta series. To do so, it will be necessary to use the following
technical
Proposition 26. Let f ∈ V gen(Zp,Γ(pn)arith) and let (Ea, ϕ, βn) be the trivialized Γ(pn)arith
elliptic curve dened over Op∞ which was attached to the couple (a, ϕ) in the previous
section. Then, letting (Ea, ϕ, β0) denote the trivialized Γ(1)arith elliptic curve attached to
(a, ϕ) and letting f (0) be the level Γ(1)arith p-adic modular form corresponding to f under
the canonical isomorphism of Proposition 18, one has the following equalities:
f (0)(Ea, ϕ, β0) = V np f(Ea, ϕ, βn) = Frobnp (f(Ea, ϕ, βn)),
where Frobp =(H(p∞)/K
p
)is the Artin symbol.
Proof. This is [Kat76, Lem.8.3.25].
It follows from Lemma 6 and the fact that
DNVp = pVpDN ,
as operators on V gen(Zp,Γ(N)arith), that
Dk1E
[p]2 = (1− pk(1 + p)Vp + p2k+1V 2
p )Dk1E2 = (1− pkVp)(1− pk+1Vp)D
k1E2.
Using the denition of
Lµ[p]E
((a, ϕ), k),
110
and the previous proposition, it then follows as in the proof of Theorem 18 that
Frob2p(Lµ[p]E
((a, ϕ), k))
Ωp(a)2k+2= (1− pk Frobp)(1− pk+1 Frobp)
(δk2E2(a)
((2πi)−1ΩC(a))2k+2
)or equivalently
Lµ[p]E
((a, ϕ), k)
Ωp(a)2k+2= (Frob−1
p −pk)(Frob−1p −pk+1)
(δkE2(a)
((2πi)−1ΩC(a))2k+2
)(8.8)
Finally, one can prove that the Petersson inner product of theta series of weight greater
than one can be p-adically interpolated, at least when K has genus one.
Theorem 19. Let K be an imaginary quadratic eld of genus one with Hilbert class eld
H and let a and b be two fractional ideals which are such that
abc2 = OK .
Let also θa,` and θb,` be the theta series of weight 2`+ 1 attached to a and b for some ` > 0.
Then for any prime p > 3 which splits in K, say pOK = pp, and any isomorphism
ϕ : Qp/Zp −→⋃n≥1
p−nc/c,
there exists a p-adic analytic function
F :W −→ Cp
with the property that
F (`)
Ωp(c)4`= (Frob−1
p −p2`−1)(Frob−1p −p2`)
(〈θa,`, θb,`〉
((2πi)−1ΩC(c))4`
)for all ` > 0,
where Frobp =(H/Kp
)is the Artin symbol.
111
Proof. Note that it follows from the theory of complex multiplication and the explicit
formulas of Proposition 13 that (〈θa,`, θb,`〉
((2πi)−1ΩC(c))4`
)∈ H,
so it makes sense to apply Frobp to it. In the light of the above computations and the
explicit formulas of Proposition 13, it suces to take
F (κ) = 4(|DK |/4)κLµ[p]E
((c, ϕ), 2κ− 1).
Note that
(|DK |/4) ∈ Z×p ,
so (|DK |/4)κ is analytic on W.
8.5 p-adic analogue of Kronecker's First Limit Formula
Recall that when ` = 0, the theta series attached to ideals in imaginary quadratic elds
are not cuspidal and so it makes no sense a priori to consider the quantity
〈θa,0, θb,0〉.
However, one can makes sense of F (0), since F is analytic on W. Using the denition of F
given in the proof of Theorem 19, one sees that
F (0) = 4Lµ[p]E
((c, ϕ),−1) = 4
∫Z×px−1dµ
[p]E (Ec, ϕ, β2).
Using the ideas of Propositions 24 and 25, one sees that∫Z×px−1dµ
[p]E = D−1
1 E[p]2 =
∑(n,p)=1
n−1σ(n)qn.
112
Our goal in this section is to nd an expression for the p-adic modular form given by the
q-expansion above in terms of known objects. It will then appear that our formula can be
seen as a p-adic analogue of Kronecker's First Limit formula.
Let g0 be the modular unit given by
g0 =∆
Vp∆, (8.9)
where ∆ ∈ S12(SL2(Z)) is the classical modular form dened in Chapter 1, whose q-
expansion is given by
∆(q) = q∏n≥1
(1− qn)24. (8.10)
Then g0 is a weakly holomorphic modular form of level Γ0(p) and weight 0. Dene its
p-depletion as
g(p)0 =
Vpg0
gp0=
(Vp∆)p+1
∆pV 2p ∆
. (8.11)
Then one has the following
Lemma 7. Let g(p)0 be as above. Then
g(p)0 ∈ 1 + pV gen
0 (Zp,Γ(p2)arith).
Proof. To see that
g(p)0 ∈ V gen
0 (Zp,Γ(p2)arith),
it suces to note that g(p)0 ∈M0(Γ0(p2)) and that it has q-expansion given by
g(p)0 (q) =
∏n≥1
((1− qpn)p+1
(1− qn)p(1− qp2n)
)24
,
which is an element of Zp[[q]]. To see that g(p)0 ∈ 1 + pV gen
0 (Zp,Γ(p2)arith), it suces to
show that
g(p)0 ≡ 1 (mod pZp[[q]]),
113
which follows directly from the fact that
(1− qpn)p+1 ≡ (1− qpn)p(1− qpn) ≡ (1− qp2n)(1− qn)p (mod pZp[[q]]).
Since the power series
log(1 + x) =∑n≥1
(−1)n+1xn
n
converges for any x ∈ pV gen0 (Zp,Γ(p2)arith), one can take the logarithm of g(p)
0 and obtain
a p-adic modular form
logp g(p)0 ∈ V gen
0 (Zp,Γ(p2)arith).
Here, logp denotes the function dened by the above power series for x ∈ pVgen
0 (Zp,Γ(p2)arith).
Then one can prove the following p-adic analogue of Kronecker's First Limit formula.
Proposition 27. In the above notation, one has that
D−11 E
[p]2 =
∫Z×px−1dµ
[p]E =
−1
24plogp g
(p)0 .
Proof. First note that the following equality holds in Zp[[q]]:
logp∏n≥1
(1− qprn)24 = 24∑n≥1
σ−1(n)qprn
114
for any integer r ≥ 0, where σ−1(n) =∑
d|n d−1. It then follows that
logp g(p)0 = −24
p∑n≥1
σ−1(n)qn − (p+ 1)∑n≥1
σ−1(n)qpn +∑n≥1
σ−1(n)qp2n
= −24p
∑n≥1
σ−1(n)qn − (1 + p−1)∑n≥1
σ−1(n)qpn + p−1∑n≥1
σ−1(n)qp2n
= −24p
∑(n,p)=1
σ−1(n)qn
= −24pD−11 E
[p]2 .
A slightly more conceptual way of proving the above result is as follows. When k ≥ 4,
rst note that the q-expansion of Ek, which is given by (1.24), can be written as
Ek(q) =ζ(1− k)
2+∑n≥1
σk−1(n)qn.
When k = 0, the above formula does not make sense because the Riemann zeta function
has a pole at s = 1. However, forgetting the constant term, one can dene formally
E0(q) =∑n≥1
σ−1(n)qn.
Then one can show that
TpE0 = (Up + p−1Vp)E0 = (1 + p−1)E0
in the same way as one shows that
TpEk = (1 + pk−1)Ek
115
when k ≥ 4. It then follows formally that
E0−(1+p−1)VpE0 +p−1V 2p E0 = E0−Vp(Up+p−1Vp)E0 +p−1V 2
p E0 = (1−VpUp)E0 = E[p]0 .
Since
E[p]0 (q) = D−1
1 E[p]2 (q)
and
logp∏n≥1
(1− qprn) = V rp E0(q),
this gives another proof of the above Proposition.
The above proposition implies that
−1
24pD1 logp g
(p)0 = E
[p]2
(compare with the computations done at the end of Section 4.3) and so
F (0) = 4D−11 E
[p]2 (Ea, ϕ, β2) =
−1
6plogp g
(p)0 (Ea, ϕ, β2). (8.12)
Generally speaking, one usually hopes that evaluating a p-adic L-function outside
its range of interpolation gives an expression which is analogous in some sense to the
corresponding complex value. Think of Kubota's formula for that value at 1 of the p-adic
zeta function, for example. In the case of (8.12), the corresponding complex analogue would
be the Petersson inner product of weight one theta series attached to ideals, which is of
course not dened. However, one could still formally use the formula of Theorem 15 for
` = 0 and obtain
〈θa,0, θb,0〉 =−1
3log(N(c)6|∆(c)|)
116
(we still suppose that K has genus one and that abc2 = OK). Using the same kind of
reasoning that lead to (8.8), one can then see that
F (0) =−1
6(Frob−1
p −p−1)(Frob−1p −1) logp ∆(c)
formally (since ∆ 6∈ 1 + pV gen(Zp,Γ(1)arith)). Notice how the last equation looks like the
p-stabilized version of the one before it.
117
Part III
Computations
118
In the third and nal Part of this thesis, many computations and numerical experiments
are presented. Almost all computations were done using the PARI/GP computer algebra
system (see [PAR16]).
Using the scripts
A big eort was made to make the code written for this project easy to use for anyone
who is a little bit familiar with PARI/GP. Some functions in PARI/GP 2.9 are required,
so it is necessary to have a version of PARI/GP which is ≥ 2.9.
To download the code, one can access the url [Sim] and download manually the whole
ENT repository or simply run the command
git clone https://github.com/NicolasSimard/ENT.git
on the terminal of any machine which has the git version control system installed (see
[Git]). Once the code is downloaded, navigate to the ENT/ folder and start the PARI/GP
calculator there.
PARI/GP Session 1.
git clone https :// github.com/NicolasSimard/ENT.git
cd ENT
gp
About the PARI/GPSessions
It is assumed that every PARI/GP session is run directly from the ENT/ folder. Most
PARI/GP sessions start with the read("init.gp") command, which loads the main scripts
that were written for this project. The functions that where written by the author are
highlighted in green and in slanted form. The output of the commands are sometimes
truncated to save some space. This is indicated by the symbol [...].
119
CHAPTER 9Petersson inner product
This chapter contains examples of computations of Petersson norm of newforms.
9.1 Petersson norm of ∆
To compute the Petersson norm of the modular form ∆, one could use directly the
denition of the Petersson inner product as a double integral:
PARI/GP Session 2.
gp > default(realprecision , 50);
gp > delta(x) = eta(x,1) ^24;
gp > intnum(x = -1/2, 1/2, intnum(y = (1-x^2) ^(1/2) ,[[1],4*Pi], \
norm(delta(x+y*I))*y^10))
%1 = 1.0353620568043209223478168122251645932249079609504 E-6
The last command runs in 4 seconds. A much more ecient method consists in dening
the symmetric square L-function of ∆ and Formula (2.12).
PARI/GP Session 3.
gp > default(realprecision , 50);
gp > N = 1; k = 12;
gp > a = (n -> vector(n,i,sumdiv(i, d, \
(-1)^bigomega(d)*d^(k-1)*ramanujantau(i/d)^2)));
gp > L = lfuncreate ([a,1,[0,1,2-k],2*k-1,N^2,1]);
gp > (Pi /2*(4* Pi)^k/(k-1)!/N)^-1*lfun(lfuncreate(Ldata),k)
%1 = 1.0353620568043209223478168122251645932249079609373 E-6
120
The formula for the coecients of the Dirichlet series are taken from [Coh13, Thm.2.1].
The last command runs in a few milliseconds, nearly 100 times faster than the previous
one. Note that the time it takes to initialize the L-function may not always be negligible. It
depends, in particular, on how eciently one can compute the coecients of the Dirichlet
series.
There are many more ways of computing the Petersson norm of ∆, like Poincaré series,
Haberland's formula, periods or other formulas involving special values of Dirichlet series.
Some methods, like Haberland's formula, are even faster than the symmetric square L-
function method. For more on those methods, see [Coh13].
9.2 Petersson norm of ∆5(τ) = (η(τ)η(5τ))4
The modular form ∆5(τ) = (η(τ)η(5τ))4 is a newform of weight 2 and level Γ0(5). To
compute its Petersson norm, one could again use the denition. To do so, one needs to nd
a fundamental domain for Γ0(5) \ H. Taking the set0 −1
1 0
∪
1 0
j 1
: 0 ≤ j ≤ 4
as a set of representatives for the quotient SL2(Z)/Γ0(5), one can write the double integral
over Γ0(5) \ H as a sum of integrals over translates of the fundamental domain of SL2(Z)
by these representatives. Another method is to use (2.12) again.
PARI/GP Session 4.
gp > default(realprecision , 50);
gp > k=4; N=5; anf(n) = lfunan(lfunetaquo ([1 ,4;5 ,4]),n);
gp > avec(n) =
my(an=anf(n), ap2);
direuler(p=2,n,
if(N%p==0,
1/(1 - an[p]^2*X), /* factor at the bad primes */
121
ap2 = an[p]^2-p^(k-1);
1/(1 - ap2*X + p^(k-1)*ap2*X^2 - p^(3*(k-1))*X^3)));
gp > L = lfuncreate ([n -> avec(n) ,1,[0,1,2-k],2*k-1,N^2 ,1]);
gp > (Pi /2*(4* Pi)^k/(k-1)!/N)^-1*lfun(L,k)
%1 = 0.00087080010497869125684556081325455557786399599585989
This last computation is more than 1000 times faster than the previous one. The
expression for the Euler factors at p of the symmetric square L-function in terms of the
Fourier coecients are easily obtained directly from the denition (2.6).
9.3 Petersson norm of ∆7(τ) = (η(τ)η(7τ))3
The modular form ∆7(q) = (η(q)η(q7))3 is newform of weight 3, level Γ0(7) and cha-
racter χ−7, the character attached to the quadratic eld of discriminant −7 (given by the
Kronecker symbol).
To compute the Petersson norm of ∆7, one can use Theorem 8 or Theorem 9. For
technical reasons, we use the rst one.1
PARI/GP Session 5.
gp > default(realprecision , 50);
gp > k=3; N=7; chi(n) = kronecker(-7,n);
gp > anf(n) = lfunan(lfunetaquo ([1 ,3;7 ,3]),n);
gp > avec(n) =
my(an=anf(n), ap2);
direuler(p=2,n,
if(N%p == 0,
1/(1-p^(k-1)*X), /* factor at the bad primes */
1 The second formula involves the computation of the residue of an L-function. In general,this data is required to dene the L-function in PARI/GP. In some cases, the functionlfunrootres() can be used to nd this data numerically, but we have not been able tocompute it in this case.
122
ap2 = an[p]^2-chi(p)*p^(k-1);
1/(1 - chi(p)*ap2*X + chi(p)*ap2*p^(k-1)*X^2 - p^(3*(k-1))*X^3)));
gp > L = lfuncreate ([n -> avec(n) ,1,[0,1,2-k],2*k-1,N^2 ,1]);
gp > (Pi/2* eulerphi(N)*(4*Pi)^k/N^2/(k-1)!)^-1*( lfun(L,k)*6/7) /*Note the 6/7*/
%1 = 0.0052288338547890255263565535732901581103627066353333
Note that the one has to multiply L(Sym2 ∆7, χ−7, s) by the Euler factor
(1− p2−s)−1
at the bad prime p = 7 to obtain a nice functional equation (this explains the presence of
the factor 6/7 in the above computation). In general, nding the bad Euler factors can
be dicult. In this case, it turns out the ∆7 is also a theta series and so one can use
Proposition 4 to guess the missing Euler factor.
123
CHAPTER 10Complex multiplication
10.1 Singular moduli
In this section we illustrate the statements of Theorem 10, which is in some sense the
starting point of the theory of complex multiplication. The following calculation illustrates
the rst three points of this Theorem.
PARI/GP Session 6.
gp > read("init.gp"); default(realprecision , 500);
gp > K = bnfinit(x^2+47); reps = redrepshnf(K); hK = #reps;
gp > f = algdep(ellj(idatouhp(K,reps [1])),hK) /* Verify point 1*/
%1 = x^5 + 2257834125*x^4 - 9987963828125*x^3 + 5115161850595703125*x^2 -
14982472850828613281250*x + 16042929600623870849609375
gp > nfisincl(f,polcompositum(K.pol ,quadhilbert(K.disc))[1]) != []
%2 = 1
gp > poldegree(polcompositum(K.pol ,f)[1]) == 2*hK /* Verify point 2*/
%3 = 1
gp > f == round(prod(i=1,hK,x-ellj(idatouhp(K,reps[i])))) /* Verify point 3*/
%4 = 1
The last point of Theorem 10 could also be veried, but we didn't try to. Note that
the fact that the polynomial f is monic proves numerically that the singular values are
integral.
124
10.2 Siegel units
In this section we illustrate the statements of Theorem 11.
PARI/GP Session 7.
gp > read("init.gp"); default(realprecision , 500);
gp > phi(K,ida) = delta(idatolat(K,1))/delta(idatolat(K,idealinv(K,ida)));
gp > K = bnfinit(x^2+71); reps = redrepshnf(K); hK = #reps;
gp > f = algdep(phi(K,reps [2]) ,2*hK) /* Verify point 1*/
%1 = 68719476736*x^6 + 2785017856*x^5 + 14351421440*x^4 + 412493295*x^3 +
3503765*x^2 + 166*x + 1
gp > nfisincl(f,polcompositum(K.pol ,quadhilbert(K.disc))[1]) != [] /* phi(K,ida) \
is in H*/
%2 = 1
gp > factor(polcoeff(f,6)) /* Verify point 2*/
%3 =
[2 36]
gp > idealfactor(K,reps [2]) /* Verify point 2*/
%4 =
[[2, [0, 1]~, 1, 1, [1, -6; 1, 0]] 1]
It should also be possible to verify point 3 of Theorem 11, but we haven't tried.
In the following PARI/GPsession, we experiment with Siegel units.
PARI/GP Session 8.
gp > read("init.gp"); default(realprecision , 500);
gp > phi(K,ida) = delta(idatolat(K,1))/delta(idatolat(K,idealinv(K,ida)));
gp > siegelunit(K,ida) = complexgen(K,idealpow(K,ida ,hK))^12* phi(K,ida)^hK;
gp > K = bnfinit(x^2+31); hK = K.clgp.no; reps = redrepshnf(K); /* hK = 3 */
gp > f = algdep(siegelunit(K,reps [2]) ,2*hK) /* Siegel units are... units!*/
125
%1 = x^6 - 1590927*x^5 + [...] + 896395574769*x^2 - 1590927*x + 1
gp > nfisincl(f,polcompositum(K.pol ,quadhilbert(K.disc))[1]) != [] /* Siegel units \
belong to H*/
%2 = 1
Note that the fact that the leading and constant terms of f are both equal to 1 proves
numerically that this Siegel unit is indeed a unit.
10.3 CM values of modular forms: classically and algebraically
Recall the the Chowla-Selberg period attached to K, dened in Proposition 7 as
ΩK =1√
4π|D|
|D|−1∏n=1
Γ
(n
|D|
)χD(n)wK/(4hK)
,
could be used to algebraize the values of level one modular forms at lattices corresponding
to fractional ideals in K.
PARI/GP Session 9.
gp > read("init.gp"); default(realprecision , 500);
gp > K = bnfinit(x^2+23); Om_K = CSperiod(K.disc); reps = redrepshnf(K);
gp > factor(algdep(delta(idatolat(K,reps [2]))/Om_K ^12 ,10))
%1 =
[x + 1 1]
[262144*x^6 - 1617920*x^5 + 5304960*x^4 - 15701487*x^3 + 25590080*x^2 - 3235840*x +
262144 1]
gp > factor(algdep(delta(idatolat(K,reps [2]))/Om_K ^12 ,20 ,100)) /*Retry , see below.*/
%2 =
[x 2]
[262144*x^6 - 1617920*x^5 + 5304960*x^4 - 15701487*x^3 + 25590080*x^2 - 3235840*x +
262144 1]
126
gp > factor(algdep(E(2,idatolat(K,1))/Om_K ^2,20)) /* Suspicious output */
%3 =
[14270068201587690836857*x^20 + [...] - 116255875884396643732694 1]
gp > default(realprecision , 1000); /* Need to increase precision. */
gp > K = bnfinit(x^2+23); Om_K = CSperiod(K.disc);
gp > factor(algdep(E(2,idatolat(K,1))/Om_K ^2,20))
%4 =
[x 1]
[47952687595882920802373496471552*x^18 - 280961949079190786543261319168*x^12 -
22089242062943952187392*x^6 - 5411082280083481 1]
The degree of the polynomial output by algdep() was chosen at random, since the
only information we have about the quotients ∆(a)/Ω12K and E2(a)/Ω2
K is that they are
algebraic. To convince ourselves that the polynomial for ∆(a)/Ω12k was good, we increased
the degree and used only the rst 100 digits of the quotient in the second call to algdep().
Since both polynomials have a common factor, it seems plausible that the quotient is a root
of this common factor. For E2, the rst call to algdep() output a suspicious polynomial
(it has degree equal to 20, large height and is a sum of monomials of every degree less than
20). After doubling the precision, the polynomial becomes much simpler and one is lead
to believe that the quotient attached to E2 is a root of it. One could again double the
precision and see that the polynomial remains the same.
It was shown in Chapter 7 that one could attach a period Ωa to any fractional ideal a
of K in such a way that
f(a) · (2πiΩa)k ∈ H,
where f is any nearly holomorphic modular form of weight k and level 1. Note that here
there is control over the algebraic eld in which the numbers land.
127
PARI/GP Session 10.
gp > read("init.gp"); default(realprecision , 500);
gp > K = bnfinit(x^2+31); hK = K.clgp.no; /*hK = 3*/
gp > reps = redrepshnf(K); Om_ida = canperiod(K,reps [2]);
gp > ida = idatolat(K,reps [2]); /* Lattice attached to ideal reps [2]*/
gp > f = factor(algdep(delta(ida)*(2*Pi*I*Om_ida)^12,2*hK))
%1 =
[x 1]
[10230338448241477052305617123765768999*x^3 + [...] + 6330188685371917375801 1]
gp > f = f[2,1];
gp > nfisincl(f,polcompositum(K.pol ,quadhilbert(K.disc))[1]) != [] /*Lands in H*/
%2 = 1
gp > pol = shimuramaass(E2 ,3) /*d^3E_2*/
%3 = -48*E2^4 + 120*E4*E2^2 - 14*E6*E2 + 25*E4^2
gp > z = substvec(pol ,[E2 ,E4,E6],[E(2,ida),E(4,ida),E(6,ida)]); /*d^3E_2(ida)*/
gp > f = algdep(z*(2*Pi*I*Om_ida)^8,2*hK)
%4 = 4182444282330998785530422041377681*x^3 - [...] - 2480023220943452566129
gp > nfisincl(f,polcompositum(K.pol ,quadhilbert(K.disc))[1]) != [] /*Lands in H*/
%5 = 1
The period Ωa, returned by the function canperiod(), is computed by rst nding
an elliptic curve dened over H isomorphic to C/a over C (this can be done explicitly by
cooking up a Weierstrass equation out of the value j(a) ∈ H) and then calling the built-in
function ellperiods on this elliptic curve.
As one can see, by using the period Ωa, one trades the explicit Chowla-Selberg formula
for more information about the algebraicity of the values.
128
CHAPTER 11Petersson inner product of theta series
11.1 Hecke characters
One advantage of Hecke characters of type A0 (as opposed to general Hecke characters)
is that they are easy to compute with. To see this, let K be an imaginary quadratic eld,
let a1, . . . , ad be a set of generators of ClK , let oi be the order of ai in ClK , let a be any
fractional ideal of K and let ψ be a Hecke character of innity type T = (k1, k2). Then
ψ(ai)oi = ψ(aoii ) = ψ(αi) = αk1i αi
k2 ,
where aoii = (αi)OK , so
ψ(ai) = ζoiαk1/oii αi
k2/oi
for some othi root of unity ζoi and so
ψ(ai) = exp(2πici/oi)αk1/oii αi
k2/oi
for some 0 ≤ ci < oi. Writing
a = µd∏i=1
aeii ,
for some 0 ≤ ei < oi and µ ∈ K×, it follows that
ψ(a) = µk1 µk2 exp
(2πi
∑i=1
eici/oi
)d∏i=1
αk1ei/oii αi
k2ei/oi .
Therefore, ψ is completely determined by its innity type and the tuple c = (c1, . . . , cd).
Conversely, it is clear that any such sequence and innity type (satisfying a certain parity
condition) denes a Hecke character. For this reason, such a Hecke character is represented
129
by the tuple
[c, T ]
in the ENT/ repository.
PARI/GP Session 11.
gp > read("init.gp");
gp > K = bnfinit(x^2+17*23*47); qhcdata = qhcinit(K);
gp > K.clgp.cyc /* order of cyclic components of Cl_K*/
%1 = [24, 2, 2]
gp > qhc = [[15 ,1 ,0] ,[8 ,0]]; /*some Hecke character of oo-type (8,0)*/
gp > p11 = idealprimedec(K,11) [1]; /*a prime of K above 11*/
gp > qhceval(qhcdata ,qhc ,p11)
%2 = 12495.16962[...] - 7630.83331[...]*I
gp > factor(round(norm (%))) /*The above number has norm 11^8, as expected */
%3 = [11 8]
11.2 Theta functions
To compute the q-expansion of the theta function attached to a Hecke character, one
could use the built-in PARI/GPfunction ideallist, which returns a list of all ideals of
norm less than a bound. However, it turns out to be more ecient to compute θa,` for each
a in a set of representatives for ClK and then use the identity in Proposition 9.
PARI/GP Session 12.
gp > read("init.gp"); default(realprecision , 5000); default(seriesprecision ,1000);
gp > K = bnfinit(x^2+13*31); qhcdata = qhcinit(K);
gp > K.clgp.cyc /* order of cyclic components of Cl_K*/
%1 = [2]
130
gp > qhc = [[0] ,[4 ,0]]; /*some Hecke character */
gp > L = ideallist(K,default(seriesprecision));
gp > F(qhc)=q*Ser(apply ((L->sum(i=1,#L,qhceval(qhcdata ,qhc ,L[i]))),L),q);
gp > F(qhc); /*few seconds */
gp > bintheta(K,qhc); /*few milliseconds */
gp > vecmax(abs(Vec(F(qhc)-bintheta(K,qhc)))) /*the two q-exp are equal*/
%2 = 1.5223063546968858611 E-5003
11.3 Petersson inner product of theta functions
Of course, the formulas used in Chapter 9 could be used to compute the Petersson
norm of theta series. However, it is often more ecient to use the formulas
〈θψ, θψ〉 =4hKwK
√|D|Γ(2`+ 1)
(4π)2`+1L(ψ2, 2`+ 1) (11.1)
and
〈θψ, θψ〉 = (|D|/4)`4hKw2K
hK∑j=1
ψ2(aj)δ2`−12 E2(aj) (11.2)
of Proposition 12 and Theorem 15, respectively. To see this, take again the eld K =
Q(√−7). Then the subspace of theta functions in S3(Γ0(7), χ−7) is one-dimensional and
θOK = 2θψ,
where ψ is the unique Hecke character of K of innity type (2, 0). A trivial bound for the
dimension of S3(Γ0(7), χ−7) as a C-vector space is 3 (see [RBvdG+08, Prop.3]) and so it
follows from a q-expansion calculation that
∆7 = θψ.
PARI/GP Session 13.
gp > read("init.gp"); default(realprecision , 100); default(timer ,1);
131
gp > K = bnfinit(x^2+7); hK = K.clgp.no; ell = 1; qhc = [[] ,[2*ell ,0]];
gp > Lsym2 = lfuncreate(lsym2data(K,qhc));
gp > Lpsi2 = lfuncreate(qhcLdata(K,2* qhc));
time = 16 ms.
gp > (2* ell)!*abs(K.disc)/(4*Pi)^(2* ell+1)*2/Pi*lfun(Lsym2 ,2*ell+1)
time = 1,607 ms.
%5 = 0.005228833854789025526356553573290158110362706635333268951985345514[...]
gp > 4*hK/K.tu[1]* sqrt(abs(K.disc))*(2* ell)!/(4*Pi)^(2* ell+1)*lfun(Lpsi2 ,2*ell +1)
time = 15 ms.
%6 = 0.005228833854789025526356553573290158110362706635333268951985345514[...]
gp > pnorm(pipinit(K),qhc)
%7 = 0.005228833854789025526356553573290158110362706635333268951985345514[...]
Compare the results of these computations with those of the PARI/GPSession 5.
Note that the formula using the Hecke L-function is about 100 times faster that the
one using the symmetric square L-function. The reason for this is that the symmetric
square L-function has degree 3, while the Hecke L-function has degree 2. Note also that
the function pnorm, which uses (11.2), runs in a negligible amount of time.
The main advantage of (11.2) is that the derivatives of E2 can be expressed recursively
in terms of E2, E4 and E6 (see (1.73), (1.74) and (1.75)), and so the evaluation of δ2`−12 E2(a)
boils down to the evaluation of a 3 variable polynomial (which depend only on `) at the
point (E2(a), E4(a), E6(a)) ∈ C3. The only problem is that the degree and the height of that
polynomial grows with ` and it becomes expensive to compute. However, since it depends
only on ` it can be computed in advance and stored. The large use of computation power
132
is then transferred to a large use in memory.1 This leads to the search for yet another
algorithm which would scale well with `. An idea is to compute explicitly the "q-expansion"
of δn2E2 using the formula
δn2E2(τ) =n∑r=0
(−1)n−r(n
r
)(r + 2)n−r
(4π=(τ))n−rDrE2(τ),
where (a)d = a(a+ 1) . . . (a+ d− 1) is the Pochhammer symbol and
D =1
2πi
∂
∂z
(see [RBvdG+08, Eqn.56]). A calculation then shows that
δn2E2(τ) =(−1)n(
1
8π=(τ)− n+ 1
24
)n!
(4π=(τ))n
+∑m≥1
σ(m)
(n∑r=0
(−1)n−r(n
r
)(r + 2)n−r
(4π=(τ))n−rmr
)qm.
(11.3)
The running time of all those formulas can be compared for varying ` and precision. Here is
a small comparison of the timing of the computation of the Petersson norm of θψ for some
Hecke character ψ of K = Q(√−47) of innity type (2`, 0) using 4 dierent algorithms at
1000 digits of precision. The timings are in seconds.
`
1 25 50 100
Eq. (11.1) 5.3 9.2 14.9 31.0
AlgorithmsEq. (11.2) 0 0.2 1.7 25.5
Eq. (11.3) 0.4 2.5 6.3 18.8
Eq. (11.3)' 0.4 0.6 0.7 1.2
1 For example, the rst 220 polynomials δ12E2, . . . , δ
4372 E2 take about 900 Mb of space.
133
Here, equation (11.3)' is a vectorized implementation of the formula (11.3), i.e. it uses
a dierent algorithm to evaluate the q-expansion. Here is the same comparison, but with
2000 digits of precision.
`
1 25 50 100
Eq. (11.1) 23.3 35.3 50.7 84.1
AlgorithmsEq. (11.2) 0 0.3 2.6 31.2
Eq. (11.3) 1.9 8.4 18.5 47.4
Eq. (11.3)' 1.8 2.4 2.8 4.1
As expected, equation (11.2) behaves well when the precision increases, but behaves
badly when ` increases. One also sees that the vectorized implementation of (11.3) does
consistently well.
One advantage of having more than one formula to compute a quantity numerically is
that one can have an idea of the number of correct digits of the results by comparing the
output of two dierent algorithms. Here is an example.
PARI/GP Session 14.
gp > default(realprecision ,1000); read("init.gp");
gp > K = bnfinit(x^2+53); pipdata = pipinit(K); K.clgp
%4 = [6, [6], [[3, 2; 0, 1]]]
gp > pnorm(pipdata ,[[1] ,[10 ,0]])-pnorm(pipdata ,[[1] ,[10 ,0]] ,"qexpv")
%5 = 1.8923323355184722792 E-991 + 8.480963825248301816 E -1006*I
The rst call to pnorm uses (11.2), while the second uses the vectorized implementation
of (11.3). The dierence between the two quantities suggest that around 990 out of the
1000 digits of precision are correct.
134
Finally, in the next PARI/GPsession, we verify (4.14), which gives an expression for
the value at s = 1 of Hecke L-functions of non-trivial class characters.
PARI/GP Session 15.
gp > read("init.gp");
gp > K = bnfinit(x^2+47); wK = K.tu[1]; reps = redrepshnf(K);
gp > qhc = [[2] ,[0 ,0]]; /*some class character */
gp > Psi(ida) = qhceval(qhcinit(K),qhc ,ida);
gp > Lpsi2 = lfuncreate(qhcLdata(K,2* qhc));
gp > F(ida) = idealnorm(K,ida)^6* abs(delta(idatolat(K,ida)));
gp > lfun(Lpsi2 ,1)
%1 = 0.64666083128645259893546663118873725573 - 7.4[...] E-60*I
gp > -Pi/3/wK/sqrt(abs(K.disc))*sum(i=1,#reps ,Psi(reps[i])^-2*log(F(reps[i])))
%2 = 0.64666083128645259893546663118873725572 + 0.E-38*I
135
CHAPTER 12Stark's observation on weight one theta functions
12.1 Stark's original observation in the eld Q(√−23)
Stark noticed that if θψ is the theta function attached to a non-trivial class character
of K = Q(√−23), then
〈θψ, θψ〉 = 3 log ε,
where ε is a real root of x3 − x− 1. This is easy to verify numerically.
PARI/GP Session 16.
gp > read("init.gp"); default(realprecision , 500);
gp > Phi(K,ida) = my(L=idatolat(K,ida)); \
sqrt(idealnorm(K,ida))*abs(L[1]^ -1* eta(L[2]/L[1] ,1)^2);
gp > K = bnfinit(x^2+23); data = pipinit(K);
gp > qhc = [[1] ,[0 ,0]]; /*A non -trivial class character */
gp > algdep(exp(pnorm(data ,qhc)/3) ,3)
%1 = x^3 - x - 1
gp > p31 = idealprimedec(K,31) [1]; /*31 splits in K*/
gp > abs(pnorm(data ,qhc) -3*log(Phi(K,idealinv(K,p31))/Phi(K,1)))
%2 = 2.583511420128310153 E-500
12.2 Generalizing Stark's observation to class number 3 quadratic elds
One may wonder if the Petersson norm of theta functions is always the logarithm of a
unit in the Hilbert class eld. This was proven to be the case for all class number 3 number
elds and can be seen numerically.
136
PARI/GP Session 17.
gp > read("init.gp"); default(realprecision , 1000);
gp > L = discofclassno (3); /*There are 16 of them*/
gp > test(K) = polredbest(algdep(exp(pnorm(pipinit(K) ,[[1] ,[0 ,0]]) /3) ,3));
gp > for(i=1,#L,print("D=",L[i],": ",test(bnfinit(x^2-L[i]))))
D=-907: x^3 - x^2 - 7*x + 12
D=-883: x^3 - 2*x^2 - 12*x - 11
D=-643: x^3 - 2*x - 5
D=-547: x^3 - x^2 - 3*x - 4
D=-499: x^3 + 4*x - 3
D=-379: x^3 - x^2 + x - 4
D=-331: x^3 - x^2 + 3*x - 4
D=-307: x^3 - x^2 + 3*x + 2
D=-283: x^3 + 4*x - 1
D=-211: x^3 - 2*x - 3
D=-139: x^3 - 8*x - 9
D=-107: x^3 - x^2 + 3*x - 2
D=-83: x^3 - x^2 + x - 2
D=-59: x^3 + 2*x - 1
D=-31: x^3 + x - 1
D=-23: x^3 - x - 1
gp > for(i=1,#L,print("D=",L[i],": ",polredbest(quadhilbert(L[i]))))
D=-907: x^3 - x^2 - 7*x + 12
D=-883: x^3 - 2*x^2 - 12*x - 11
D=-643: x^3 - 2*x - 5
D=-547: x^3 - x^2 - 3*x - 4
D=-499: x^3 + 4*x - 3
D=-379: x^3 - x^2 + x - 4
D=-331: x^3 - x^2 + 3*x - 4
D=-307: x^3 - x^2 + 3*x + 2
D=-283: x^3 + 4*x - 1
D=-211: x^3 - 2*x - 3
D=-139: x^3 - 8*x - 9
D=-107: x^3 - x^2 + 3*x - 2
137
D=-83: x^3 - x^2 + x - 2
D=-59: x^3 + 2*x + 1
D=-31: x^3 + x - 1
D=-23: x^3 - x - 1
12.3 Generalizing Stark's observation to other quadratic elds
In this section, we give examples of computations which support the claims and con-
jectures made in Section 5.3. Recall the notation
〈θψ, θψ〉 = hK log κψ,
where
κψ =∏
a∈ClK
Φ(a−1)−ψ(a)2 .
PARI/GP Session 18.
gp > read("init.gp"); default(realprecision , 1000);
gp > Phi(K,ida) = my(L=idatolat(K,ida)); \
sqrt(idealnorm(K,ida))*abs(L[1]^ -1* eta(L[2]/L[1],1)^2);
gp > k_psi(K,qhc) =
my(reps=redrepshnf(K));
prod(i=1,K.clgp.no,Phi(K,reps[i])^(-qhceval(qhcinit(K),qhc ,reps[i])));
gp > Q(D) = bnfinit(x^2-D); /*for convenience */
gp > algdep(k_psi(Q(-47) ,[[1] ,[0 ,0]]) ,50) /*Class number 5: hummm*/
%1 = 388601289028*x^50 - [...] - 6975618148752847
gp > factor(algdep(k_psi(Q(-87) ,[[1] ,[0 ,0]]) ,50)) /*Class number 6: it is a unit!*/
%2 =
[ x - 1 1]
[x^6 - x^5 - 6*x^4 - 17*x^3 - 6*x^2 - x + 1 1]
gp > nfisincl (%[2,1], polcompositum(Q(-87).pol ,quadhilbert (-87))[1]) != [] /*Is in \
H*/
138
%3 = 1
gp > D10 = discofclassno (10) [ -1.. -1][1]; /*Class number 10*/
gp > algdep(k_psi(Q(D10) ,[[1] ,[0 ,0]]) ,50) /* Garbage for this Hecke char*/
%4 = 387398835037687*x^50 + [...] + 4477373333556254
gp > algdep(k_psi(Q(D10) ,[[5] ,[0 ,0]]) ,50) /*Works with this Hecke char*/
%5 = x^31 - 8*x^30 - x^29
Many other experiments like these where run and they all support the conjectures
made in Chapter 5.
139
CHAPTER 13Computational experiments and conjectures
In this last chapter of the thesis, we present some computational experiments which
lead to some conjectures. For most of these conjectures, we have only numerical evidence.
13.1 Gram matrix of the Petersson inner product on the space of theta series
If f1, . . . , fn ∈ Sk(Γ0(N), χ) are cusp forms, dene
Gram(f1, . . . , fn) = det(〈fi, fj〉)1≤i,j≤n.
This is the determinant of the Gram matrix of the space spanned by f1, . . . , fn and equipped
with the Petersson inner product.
When ` > 0, recall that by Proposition 10 the space
ΘK,` ⊆M2`+1(Γ0(|D|), χD)
of theta series has dimension hK . It is then natural to consider
Gram(θψ1 , . . . , θψhK ),
where the θψi are the theta series attached to Hecke characters of innity type (2`, 0). This
quantity depends only on K and ` and by Corollary 5, it is an algebraic multiple of Ω4`hKK ,
where ΩK is the Chowla-Selberg period attached to K. In fact, numerically we nd that
this algebraic number, denoted
A(K, `) =Gram(θψ1 , . . . , θψhK )
Ω4`hKK
=
∏hKi=1〈θψi , θψi〉
Ω4`hKK
, (13.1)
140
is almost always an integer (some small powers of 2 and 3 appear in the denominator
sometimes when ` = 1).
In the following PARI/GP session, the function invA returns A.
PARI/GP Session 19.
gp > read("init.gp"); default(realprecision , 500);
gp > ell = 1; for(D = 5, 50, if(isfundamental(-D),print(-D,": \
",algdep(invA(-D,ell) ,5))));
-7: 3*x^5 - x^4
-8: 2*x - 1
-11: x - 1
-15: x - 4
-19: 3*x^5 - 13*x^4
-20: x - 64
-23: x - 621
-24: x - 196
-31: x - 7254
-35: x - 324
-39: x - 4076800
-40: x - 3364
-43: 3*x - 214
-47: x - 538443750
gp > ell = 2; for(D = 5, 50, if(isfundamental(-D),print(-D,": \
",algdep(invA(-D,ell) ,5))));
-7: x - 1
-8: x - 5
-11: x - 10
-15: x - 6084
-19: x - 142
-20: x - 21904
-23: x - 7303581
-24: x - 318096
-31: x - 404717958
141
-35: x - 5702544
-39: x - 16446807606528
-40: x - 36820624
-43: x - 22588
-47: x - 480609496085043750
This observation about the rationality of the normalized product of special values of
Hecke L-functions in (13.1) is similar in nature to what was observed by Gross and Zagier
in [GZ80] or by Villegas and Zagier in [VZ92]. In those papers, they observe (and prove
in certain cases), that the central critical value of Hecke characters attached to CM elliptic
curves is an integral multiple of powers of the corresponding Chowla-Selberg period, up to
some explicit factors.
One strategy to prove this rationality property would be to introduce a Galois action
and to show that the product in (13.1) is Galois invariant. The following experimentation
explores this idea.
PARI/GP Session 20.
gp > read("init.gp"); default(realprecision , 2000);
gp > K = bnfinit('x^2+23); data = pipinit(K); Om_K = CSperiod(K.disc);
gp > ell = 1; qhcs = qhchars(K,[2*ell ,0]);
gp > for(i = 1, K.clgp.no , print(algdep(pnorm(data ,qhcs[i])/Om_K ^(4* ell) ,10)))
x^9 - 6966*x^6 + 11569230*x^3 - 239483061
x^9 - 6966*x^6 + 11569230*x^3 - 239483061
x^9 - 6966*x^6 + 11569230*x^3 - 239483061
gp > ell = K.clgp.no; qhcs = qhchars(K,[2*ell ,0]);
gp > for(i = 1, K.clgp.no , print(algdep(pnorm(data ,qhcs[i])/Om_K ^(4* ell) ,10)))
x - 5055
x^6 - 16287872873193*x^3 + 30021979248651078296845875
x^6 - 16287872873193*x^3 + 30021979248651078296845875
142
Many observations can be made after those computations. First, one sees that the
quantities
N(ψ, `) =〈θψ, θψ〉
Ω4`K
are Galois conjugate over Q when ` is not a multiple of hK (this phenomenon repeats not
only when ` = 1). Moreover, the N(ψ, `) are hKth roots of an algebraic integer. Also,
there is this interesting phenomenon that when ` is a multiple of hK , one of the N(ψ, `)
is rational and the other two are Galois conjugate over Q. Similar observations were made
for all the number elds tested (including some number elds with more than one genera).
The last observation suggests that one Hecke character is singled out when ` is a
multiple of the class number, which is surprising at rst. However, one sees that when
hK |`, the map
ψ0,`(a) = α2`/hK ,
where α is a generator of ahK is a well-dened Hecke character of innity type (2`, 0). This
ψ0,` is the singled out Hecke character. In the notation of Section 11.1, it corresponds to
the tuple c = (0, . . . , 0). Note that when ` = 0, this is just the trivial characters. It would
be interesting to see what can be said about the special values of the Hecke L-function
corresponding to this special Hecke character.
13.2 Invariant theta series attached to imaginary quadratic elds
When ` > 0, one can also compute the determinant of the Gram matrix of the Petersson
inner product on the space ΘK,` with respect to a basis consisting of theta series attached
to fractional ideals of K. If a1, . . . , ahK is a choice of representatives of ClK , one could
choose the basis of ΘK,` to be
θa1,`, . . . , θahK ,`.
143
The problem is that
Gram(θa1,`, . . . , θahK ,`)
depends on this choice. Indeed,
Gram(θa1,`, . . . , θµai,`, . . . , θahK ,`) = N(µ)2` Gram(θa1,`, . . . , θai,`, . . . , θahK ,`) forall µ ∈ K×.
The obvious way to solve this problem is to normalize the Gram determinant using the
norm function. This leads to the quantity
B(K, `) =Gram(θa1,`, . . . , θahK ,`)
Ω4`hKK
∏hKi=1N(ai)2`
,
an algebraic number which depends only on K and `. Numerically, one notices that B(K, `)
is rational. In fact, this follows from the observation that A(K, `) is rational, because of
the following
Proposition 28. Let A(K, `) and B(K, `) be the quantities dened above. Then
B(K, `) =
(w2K
hK
)hKA(K, `).
Note that the proportionality factor does not depend on `.
Proof. Let
B1 = θa1,`, . . . , θahK ,`
and
B2 = θψ1 , . . . , θψhK
be bases for the space ΘK,`. Using Proposition 9, one computes that
Gram(θa1,`, . . . , θahK ,`) = det(T t) Gram(θψ1 , . . . , θψhK ) det(T ), (13.2)
144
where T is the hk × hk matrix whose ij entry is wkhKψi(aj), the bar denotes complex conju-
gation and the t denotes transposition.
Now
det(T ) =
(wKhK
)hK hk∏j=1
ψ1(aj)
det
(ψiψ1
(aj)
).
As i goes from 1 to hK , the class character χi = ψi/ψ1 goes through all the class characters
exactly once. Letting M be the matrix with ij entry χi(aj), one then sees that
det(T tT ) =
(wKhK
)2hK(hk∏i=1
|ψ1(ai)|2)
det(M tM) =
(wKhK
)2hK(hk∏i=1
N(ai)
)det(M tM).
Now the ij entry of M tM is
hK∑k=1
χk(ai)χk(aj) = hKδi,j ,
so that
det(M tM) = hhKK .
Putting everything together, one has
Gram(θa1,`, . . . , θahK ,`) =
(w2K
hK
)hK ( hk∏i=1
N(ai)
)Gram(θψ1 , . . . , θψhK ), (13.3)
which proves the claim.
Note also that by Corollary 7, one has the equality
Gram(θa1,`, . . . , θahK ,`)∏hKi=1N(ai)2`
= det(〈θOK ,`, θa−1
i aj ,`〉)
1≤i,j≤hK.
Although the computations with the invariantA(K, `) seem to suggest that the Chowla-
Selberg is a good choice of period to normalize the Petersson inner product of theta series,
it would be nice to nd a way to get rid of this choice of period. In what follows, we do
this when DK < −4.
145
First, dene
θA,` =θa,`
E2(a−1)`,
where A ∈ ClK and a is a representative of A.1 Then the set θA1,`, . . . , θAhK ,`, where
ClK = A1, . . . ,AhK is a basis of ΘK,` with the property that
〈θAi,`, θAj ,`〉 ∈ H
for any 1 ≤ i, j ≤ hK by the theory of complex multiplication. Note that no choice of
period is involved in the previous statement. Moreover, note that when ` = 0, we recover
the usual theta series θA,0.
Dene
C(K, `) = Gram(θA1,`, . . . , θAhK ,`).
Then
C(K, `) =Gram(θa1,`, . . . , θahK ,`)∏hK
i=1 |E2(a−1i )|2`
,
where a1, . . . , ahK is any choice of class representative for ClK . Numerically, one notices
that C(K, `) is rational whose denominator is a 2`th power of a xed number (as usual,
up to powers of 2 or 3). Here is a table of the denominators that appear for class number
1, 2, 3 and 4 and small values of `
1 The reason for the exclusion of the imaginary quadratic elds of discriminant −3 and−4 is that E2 vanishes at the CM point corresponding to them.
146
K
Q(√−163) Q(
√−187) Q(
√−23) Q(
√−203)
`
1 2 · 3 · 1812 24132 4192 24572592
2 241814 210134 4194 212572594
3 2101816 218136 4196 218572596
4 281818 210138 4198 572598
5 21218110 2261310 41910 2325725910
6 21918112 2361312 41912 2405725912
7 21918114 2401314 41914 2525725914
8 22018116 2421316 41916 2525725916
9 22318118 2441318 41918 2465725918
10 22518120 2521320 41920 2645725920
Those numbers also appear in a relation between the invariants B and C. Indeed, it
seems numerically that
B(K, `) =n(K)2`
(122|D|)hK`C(K, `),
where n(K) is the number such that n(K)2` is the denominator of C(K, `). Of course, this
conjecture is not very precise, but hopefully the following computations might convince the
reader that it makes sense.
PARI/GP Session 21.
gp > read("init.gp"); default(realprecision , 1000);
gp > D = -23; hK = qfbclassno(D)
%1 = 3
gp > for(ell = 1, 10, \
print(ell ,":",factor(invB(D,ell)/invC(D,ell)*(12^2* abs(D))^(hK*ell))))
1:Mat([419 , 2])
147
2:Mat([419 , 4])
3:Mat([419 , 6])
4:Mat([419 , 8])
5:Mat([419 , 10])
6:Mat([419 , 12])
7:Mat([419 , 14])
8:Mat([419 , 16])
9:Mat([419 , 18])
10: Mat([419, 20])
gp > default(realprecision , 2000); D = -95; hK = qfbclassno(D)
%1 = 8
gp > for(ell = 1, 10, \
print(ell ,":",factor(invB(D,ell)/invC(D,ell)*(12^2* abs(D))^(hK*ell))))
1:[1531 , 2; 242798651 , 2]
2:[1531 , 4; 242798651 , 4]
3:[1531 , 6; 242798651 , 6]
4:[1531 , 8; 242798651 , 8]
5:[1531 , 10; 242798651 , 10]
6:[1531 , 12; 242798651 , 12]
7:[1531 , 14; 242798651 , 14]
8:[1531 , 16; 242798651 , 16]
9:[1531 , 18; 242798651 , 18]
10:[1531 , 20; 242798651 , 20]
It would be interesting to see if the numbers n(K) can be related to K in some other
way.
148
Conclusion
As was seen in this thesis, Stark's observation on the relation between the Petersson
norm of the weight one theta series attached to a non-trivial character of K = Q(√−23)
and the logarithm of a unit can sometimes be generalized to other number elds. It would
be interesting to see if Conjecture 1 holds. In any case, it was shown using the explicit
formulas of Theorem 15 that the Petersson norm of weight one theta series attached to
class characters of imaginary quadratic elds is a linear combination of logarithms of Siegel
units.
In the second part, it was shown that the Petersson inner product of higher weight
theta series can be p-adically interpolated. At the point corresponding to weight one, which
was outside the range of interpolation, it was shown that one obtains a p-adic analogue of
the Petersson inner product of weight one theta series (if it could be dened).
The possibility of eciently computing the Petersson inner product of theta series
allows one to experiment with them and observe some interesting patterns. Some expe-
riments were presented in the last chapter. It would be interesting to further investigate
the many open questions that were made there. In particular, the relation between the
invariant C(K, `) and the Chowla-Selberg period seems worthy of interest.
Another direction for further research would be to study the Petersson inner product
of theta series attached to type A0 Hecke characters with non-trivial conductor. Those
theta series share many common properties with the ones considered in this thesis and all
of the steps in nding the explicit formulas of Theorem 15 should generalize.
149
REFERENCES
[Coh07] H. Cohen, Number theory: Volume II: Analytic and modern tools, GraduateTexts in Mathematics, Springer New York, 2007.
[Coh13] , Haberland's formula and numerical computation of Petersson scalarproducts, The Open Book Series 1 (2013), no. 1, 249270.
[Col04] Pierre Colmez, Fontaine's rings and p-adic L-functions, http://staff.ustc.edu.cn/~yiouyang/colmez.pdf, 2004.
[DS87] E. De Shalit, Iwasawa theory of elliptic curves with complex multiplication:P-adic L-functions, Perspectives in mathematics, Academic Press, 1987.
[DS05] F. Diamond and J. Shurman, A rst course in modular forms, Graduate Textsin Mathematics, Springer, 2005.
[Git] git version control system, https://git-scm.com/.
[GZ80] Benedict H Gross and Don Zagier, On the critical values of Hecke L-series,Mémoires de la Société Mathématique de France 2 (1980), 4954.
[Hid81] Haruzo Hida, Congruences of cusp forms and special values of their zetafunctions., Inventiones mathematicae 63 (1981), 225262.
[Iwa97] H. Iwaniec, Topics in classical automorphic forms, Graduate studies in mat-hematics, American Mathematical Society, 1997.
[Kan12] Ernst Kani, The space of binary theta series., Annales des sciences mathéma-tiques du Québec 36 (2012), 501534.
[Kat76] N. M. Katz, p-adic interpolation of real analytic Eisenstein series, Annals ofMathematics 104 (1976), no. 3, 459571.
[Lan87] S. Lang, Elliptic functions, Graduate texts in mathematics, Springer, 1987.
[MM06] T. Miyake and Y. Maeda, Modular forms, Springer Monographs in Mathe-matics, Springer Berlin Heidelberg, 2006.
[MSD74] Barry Mazur and Peter Swinnerton-Dyer, Arithmetic of Weil curves, Inven-tiones mathematicae 25 (1974), no. 1, 161.
150
151
[PAR16] The PARI Group, Univ. Bordeaux, PARI/GP version 2.9.0, 2016, availablefrom http://pari.math.u-bordeaux.fr/.
[RBvdG+08] K. Ranestad, J.H. Bruinier, G. van der Geer, G. Harder, and D. Zagier, The 1-2-3 of modular forms: Lectures at a summer school in Nordfjordeid, Norway,Universitext, Springer Berlin Heidelberg, 2008.
[Shi71] G. Shimura, Introduction to the arithmetic theory of automorphic functions,Kanô memorial lectures, Princeton University Press, 1971.
[Shi75] , On the holomorphy of certain Dirichlet series, Proceedings of theLondon Mathematical Society s3-31 (1975), no. 1, 7998.
[Shi76] , The special values of the zeta functions associated with cusp forms,Communications on Pure and Applied Mathematics 29 (1976), no. 6, 783804.
[Shi10] , Elementary Dirichlet series and modular forms, Springer Mono-graphs in Mathematics, Springer New York, 2010.
[Sil94] J.H. Silverman, Advanced topics in the arithmetic of elliptic curves, Graduatetexts in mathematics, Springer-Verlag, 1994.
[Sil09] , The arithmetic of elliptic curves, Graduate Texts in Mathematics,Springer, 2009.
[Sim] N. Simard, Nicolas Simard's github ent repository, https://github.com/
NicolasSimard/ENT.
[Sta75] H. M. Stark, L-functions at s = 1. II. Artin L-functions with rational charac-ters, Advances in Mathematics 17 (1975), no. 1, 60 92.
[VZ92] F.R. Villegas and D. Zagier, Square roots of central values of Hecke L-series.
[Wat04] Mark Watkins, Class numbers of imaginary quadratic elds, Mathematics ofComputation 73 (2004), no. 246, 907938.